Integrating image processing with deep convolutional neural networks for gene selection and cancer classification using microarray data PDF Free Download
1 / 23/23
100%
Integrating image processing
with deep convolutional neural
networks for gene selection
and cancer classication using
microarray data
Yuanyuan Zhang1, Jing Chen1 & Chong Zhang2
Microarray technology has revolutionized cancer genomics by enabling the simultaneous analysis
of thousands of gene expressions, providing critical insights into gene regulation and disease
mechanisms. However, the inherent challenges of high-dimensionality, noise, and sparsity in
microarray data demand robust analytical approaches. Image processing techniques further enhance
this analysis by extracting meaningful patterns from histological and microarray-derived visual data,
aiding in biomarker discovery and classication. This study presents a novel framework leveraging
deep neural networks for gene selection and cancer classication using microarray data, addressing
the challenges of high dimensionality, noise, and sparsity. The proposed Gene-Optimized Neural
Framework (GONF) integrates the Minimum Redundancy Maximum Relevance (mRMR) gene
selection method with a deep Convolutional Neural Network (CNN) for eective feature selection and
classication. By optimizing hyperparameters and employing advanced preprocessing techniques, the
framework enhances computational eciency and accuracy. Experiments were conducted on TCGA
and AHBA datasets, utilizing metrics such as accuracy, precision and recall for evaluation. The GONF
outperformed other methods, achieving a classication accuracy of 97% on the TCGA dataset and 95%
on the AHBA dataset. The framework demonstrated signicant reductions in false positive and false
negative rates, improving cancer subtype predictions and providing biologically interpretable results.
The ndings highlight GONF’s robustness and adaptability, paving the way for its application in other
genomic studies and clinical settings.
Keywords Microarray technology, Deep neural networks, Gene selection, Cancer classication,
Convolutional neural networks, mRMR, Genomic analysis
Microarray technology has revolutionized the eld of genomics, enabling the high-throughput measurement of
gene expression across thousands of genes simultaneously1,2. is advancement has made microarray data an
invaluable resource in cancer research, oering insights into gene regulation, biological pathways, and disease
mechanisms3. A critical application of this technology is cancer classication, where accurately distinguishing
between cancer subtypes is essential for diagnosis, prognosis, and treatment planning. Despite its potential, the
analysis of microarray data presents signicant challenges due to its high dimensionality, sparsity4, and noise,
which oen obscure meaningful biological patterns and complicate the extraction of relevant information.
Gene selection plays a crucial role in addressing these challenges. By identifying the most informative genes,
gene selection reduces the dimensionality of the data, improving classication performance and enhancing the
interpretability of the results4,5. However, traditional gene selection methods, which typically rely on statistical
tests or basic feature selection algorithms, struggle to capture the complex, nonlinear relationships that exist
between genes6. ese limitations necessitate the development of more sophisticated approaches to gene
selection and classication.
1School of Medical, Technology and Information Engineering, Zhejiang Chinese Medical University, HangZhou
310053, China. 2Division of Thoracic, Surgery, the First Aliated Hospital, School of Medicine, Zhejiang University,
Hangzhou 310003, China. email: zy_zc_2002@sina.com
OPEN
Scientic Reports | (2025) 15:39465 1
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports
Deep neural networks (DNNs) have emerged as powerful tools for addressing these challenges. eir ability to
model complex, nonlinear interactions and automatically extract hierarchical features makes them particularly
well-suited for analyzing high-dimensional genomic data. DNNs can be used to enhance gene selection by
learning which genes are most relevant to the classication task, without the need for manual intervention or
predened assumptions about gene interactions7. Additionally, DNNs’ success in image processing tasks, where
they have demonstrated an ability to learn spatial hierarchies of features, suggests that similar techniques can be
applied to gene expression data to uncover hidden patterns. Integrating DNNs with gene selection techniques
oers a promising approach to improving cancer classication from microarray data. is combination allows
for the simultaneous reduction of dimensionality and the identication of biologically signicant genes, thereby
addressing issues of noise, redundancy, and computational complexity. By leveraging both the power of deep
learning and advanced feature selection, this research aims to provide a more accurate, interpretable, and
biologically meaningful analysis of cancer data8.
e motivation for this study arises from the need to develop more eective cancer classication models that
balance dimensionality reduction, interpretability, and classication accuracy. With the growing availability of
large-scale genomic data, there is an urgent need for methods that can extract meaningful insights from these
high-dimensional datasets while managing inherent noise and redundancy. By combining DNNs with robust
gene selection techniques, this research seeks to bridge the gap between cutting-edge computational methods
and their practical applications in cancer genomics. is study, makes a number of new contributions to the eld
of cancer classication using microarray data:
• Integrated mRMR-CNN Pipeline: GONF is the only program that combines the mRMR gene selection
method with a deep CNN in a single framework. GONF improves both accuracy and interpretability by using
a low-redundancy, high-relevance gene subset that directly informs the CNN’s feature extraction process.
is is better than traditional methods that treat feature selection and classication as separate steps.
• Image Processing for Genomic Data: GONF pioneers the integration of advanced image processing tech-
niques, such as the Hough Transform and Watershed segmentation, to preprocess microarray-derived visual
data and histopathological images. is preprocessing improves the quality of the input features by lowering
noise and making it easier for the CNN to nd biologically relevant patterns. is is a new method that hasn’t
been used much in previous gene selection studies.
• Optimized CNN Architecture for Genomic Data: e customized CNN architecture has six convolutional
layers, dropout regularization, and max pooling. It is made to handle the high dimensionality and sparsity
of microarray data. When combined with Random Search hyperparameter optimization, this makes GONF
stand out from other hybrid deep learning models because it ensures strong performance and computational
eciency.
ese contributions together make GONF a strong and new way to classify cancer, and it could also be useful in
other areas of genomic research and clinical diagnostics. e remainder of this article is structured as follows:
section “Related works” reviews the role of microarray technology, gene selection, and the application of deep
learning in cancer research. Section“Dataset” presents the proposed methodology, detailing the integration
of DNNs with gene selection techniques. Section“Proposed method” reports experimental results, comparing
the performance of the proposed approach with existing methods. Finally, section“Evaluation” discusses the
implications, potential applications, and future research directions.
Related works
Recent years have witnessed signicant advances in the application of neural networks for gene selection and
cancer classication using microarray data. Microarray technology generates extensive gene expression proles,
making deep learning models essential for ecient feature selection, improving classication accuracy, and
addressing challenges related to data sparsity and high dimensionality. e authors in9 introduced a DNN
methodology to enhance gene selection for cancer classication. is approach aimed to improve classication
accuracy by reducing the complexity of microarray data, integrating gene selection with a DNN for enhanced
predictive results. e ndings showed substantial improvements in predictive accuracy compared to
conventional methods, demonstrating the eectiveness of DNNs in handling high-dimensional cancer data.
Low-light image enhancement methods based on physical models of fogging have played an important role in
improving the quality of input data for feature extraction10. e focus was on minimizing noise in microarray
samples and identifying the most relevant genes for classication. is model outperformed traditional
classiers, highlighting the ability of DNNs to autonomously discern essential properties in high-dimensional
cancer data and showing their prociency in multi-class cancer classication. In11, the authors addressed the
problem of class imbalance in cancer classication by incorporating a weighted loss function within a deep
learning framework. is strategy signicantly improved the accuracy of identifying minority cancer types,
emphasizing deep learning’s capacity to manage complex real-world datasets. Multiexponential representation
learning in molecular property prediction has been introduced as a powerful approach for extracting complex
features from genomic data12. is hybrid model enhanced computational eciency and classication precision
by selecting a smaller set of genes while maintaining high predictive performance, validating that hybrid DNNs
could improve both eciency and accuracy in cancer classication.
Another study by the authors in13 explored the use of deep reinforcement learning (DRL) to improve gene
selection for cancer categorization. By progressively identifying the most important genes based on their impact
on classication accuracy, the DRL-based model reduced the number of genes needed for precise classication,
while maintaining high performance. is approach highlighted the potential of reinforcement learning in
enhancing gene selection for cancer classication. e authors in14 proposed a DNN model for predicting cancer
Scientic Reports | (2025) 15:39465 2
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
outcomes using gene expression data. e model incorporated a feature selection phase, ensuring that only the
most signicant genes were used for prediction. is model surpassed conventional classication algorithms,
oering a more ecient and accurate method for predicting cancer outcomes. A study by the authors in15
proposed a deep learning system for tumor classication, combining unsupervised training for feature extraction
with supervised training for cancer classication. is dual-phase methodology enhanced both gene selection
and classication accuracy, with the unsupervised training helping to uncover complex gene interactions,
improving classication results. Analysis of microarray data in a recent study has revealed common pathways in
gastrointestinal and cardiac diseases, conrming the importance of using microarray data in discovering disease
mechanisms16. is method signicantly reduced the number of genes required, enhancing the eciency of the
classication process while maintaining high accuracy. e study demonstrated that autoencoder-based DNN
models are eective in managing large-scale gene expression datasets. e authors in17 integrated evolutionary
methods with DNNs for feature selection and cancer classication. e emphasis was on identifying a minimal set
of genes that provided optimal classication performance. is hybrid approach achieved substantial precision,
demonstrating the eectiveness of optimization methods in enhancing cancer classication. e development
of video resolution enhancement methods using complex CNNs has shown that spatiotemporal convolutional
operations can enhance the quality of image reconstruction. is can be eective in improving the quality of
microarray-derived images18. e model adapted during training to focus on the most relevant genes, achieving
superior classication accuracy and computational eciency compared to traditional classiers. is exible
feature selection process minimized overtting and enhanced the generalizability of the model. A study by the
authors in19 employed both unsupervised and supervised learning techniques, using autoencoders for feature
extraction and DNNs for classication. is approach eectively identied pertinent gene sets and categorized
cancer types with high precision, even with limited training data. e authors in20 proposed a hybrid model
combining a deep CNN with a support vector machine (SVM) for gene selection in cancer classication. e
model’s unique architecture facilitated ecient feature extraction from microarray data, resulting in enhanced
classication accuracy. e SIR-3DCNN framework has provided good performance by combining time series
and spatial features for early detection of lung cancer, which is in line with our goal of using complex CNNs
on multidimensional data21. By representing gene interactions as a graph, GTCGA could capture complex
relationships among genes, outperforming conventional methods in both precision and interpretability.
is survey indicates that current approaches include several shortcomings. Numerous conventional models
are susceptible to overtting, especially when processing high-dimensional data, hence diminishing their
performance on novel, unexplored data. Moreover, models such as KNN are decient in sophisticated feature
selection methods, leading to suboptimal identication of essential predictors. Conventional techniques may
lack robustness, constraining the models’ capacity to reliably discern the most signicant predictors. Moreover,
although deep learning methodologies possess potential, their amalgamation with optimal feature selection
for gene selection and classication of cancer remains inadequately investigated. Resolving these diculties
necessitates the creation of models that harmonize precision with computing eciency and possess the
robustness to generalize across many situations. e proposed method GONF, overcomes many constraints of
conventional approaches by exhibiting markedly superior accuracy. is enhancement is attained by rened
selection of features and DNN optimization methods. Furthermore, the amalgamation of mRMR with deep
learning eciently diminishes data dimensionality, enabling the model to concentrate on essential variables
and mitigate overtting. e model experiences accelerated convergence owing to a random approach for
hyperparameter optimization, hence lowering training duration. e architecture facilitates scaling via transfer
learning, enhancing its adaptability to various illnesses and expanding its use in predictive analytics. e
algorithm’s ecacy and precision render it suitable for practical application, facilitating its incorporation into
healthcare environments.
e use of neural networks for gene selection and cancer classication has advanced signicantly in recent
years. Hybrid approaches such as CNN + wrapper FS22,23, mRMR + DNN24,25, CSSMO-based deep learning
(CSSMO-DL)26, and Multi-classication GAN with Feature Bundling (MGAN-FB)27 have shown promise.
Nevertheless, these techniques have drawbacks, such as the potential for overtting, the need for computationally
intensive feature selection, and the restricted ability to integrate multi-modal data, like histopathological images
or visual data derived from microarrays. For example, CNN + wrapper FS models mainly rely on numerical data
and lack robust image preprocessing, whereas mRMR + DNN approaches frequently handle feature selection
and classication in a sequential manner. Similarly, MGAN-FB concentrates on numerical data with feature
bundling, omitting CNN-specic optimization, and CSSMO-DL uses metaheuristic feature selection but
excludes visual data.
e suggested approach (GONF) lls these gaps by combining mRMR gene selection with a customized six-
layer CNN and sophisticated image processing methods like Watershed segmentation and Hough Transform
to take advantage of both visual data (microarray-derived intensity maps and histopathological images) and
numerical gene expression. Table1 compares GONF to these new hybrid methods, emphasizing how its multi-
modal data integration and unied pipeline improve accuracy and biological interpretability. is comparison
demonstrates how GONF is unique in overcoming the drawbacks of current approaches, opening the door for
its thorough methodology in section “Proposed method”.
Dataset
We evaluate the proposed method with two dierent standard microarray datasets e Cancer Genome Atlas
(TCGA) and Allen Human Brain Atlas (AHBA). Figures1 and 2 shows an example of the images used in TCGA
and AHBA datasets.
Scientic Reports | (2025) 15:39465 3
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
TCGA dataset
TCGA integrates microarray-based gene expression data with high-resolution histopathological images to
support cancer research. e microarray data provides detailed gene expression proles across thousands of
genes for various cancer types, allowing researchers to investigate transcriptional changes, identify biomarkers,
and understand tumor biology. ese datasets are complemented by clinical metadata, such as tumor stage
and patient outcomes, enabling a multi-dimensional analysis of cancer progression and response to treatments.
Alongside the microarray data, TCGA oers high-resolution histopathological Whole Slide Images (WSIs) of
tumor tissues, typically stained with Hematoxylin and Eosin (H&E). ese images allow for the examination
of tissue architecture and cellular morphology, which can be correlated with gene expression patterns derived
from microarrays. is integration of transcriptomic and imaging data supports innovative research, such as
linking molecular signatures to histological features or predicting genomic alterations from image data. e
TCGA datasets are accessible through platforms like the GDC Data Portal and the Cancer Digital Slide Archive,
making it a vital resource for advancing cancer diagnostics and therapeutics. e GCCNet model using gated
cross-correlation has shown high accuracy in multi-view classication. is idea can also be inspiring in multi-
view analysis of gene expression data28.
AHBA dataset
e AHBA combines microarray-based gene expression data with high-resolution brain imaging to map the
molecular architecture of the human brain. Microarray data includes the expression proles of over 20,000
Aspect mRMR + DNN CNN + Wrapper FS CSSMO-DL MGAN-FB Proposed GONF
Methodology
Combines
mRMR for
gene selection
with DNNs
for classication;
sequential
steps
Uses CNNs
with wrapper
FS for
iterative gene
selection
Hybrid
CSSMO
(SMO + CSA)
for FS, followed
by DL
classication
MGAN with
feature bundling
for multi-class
classication
of high-
dimensional
data
Integrates
mRMR with a
deep CNN in
a unied
pipeline,
optimizing
feature
extraction
Data
integration
Numerical
gene expression
data; no
image-based
data integration
Numerical
gene expression
data; no
image
preprocessing
Numerical
gene expression
data; no
image
integration
Numerical
gene expression
data with
feature bundling;
no image
data
Combines
numerical
gene expression
with preprocessed
histopathological and
microarray-
derived
visual data
Image
processing
None; no
preprocessing of
visual data
Minimal or no
image processing;
focuses on
numerical data
None; focuses
on numerical
gene data
None; focuses
on numerical
data with
bundling
Uses Hough
Transform,
brightness
calculations,
and Watershed
segmentation for
histopathological images
and microarray
visual data
Feature
selection
mRMR
selects low-
redundancy,
high-relevance
genes; not
iterative
during
training
Wrapper
methods
iteratively
select genes
based on
performance
CSSMO
combines
SMO and CSA,
enhanced by
mRMR for
dimensionality
reduction
Feature
bundling
reduces
dimensionality; no
explicit FS
method like
mRMR
mRMR
selects
TCGA and
AHBA genes,
optimized for
CNN feature
extraction
CNN
architecture
Generic
DNNs; not
optimized for
genomic data
sparsity
CNNs not
tailored for
sparse
genomic data
or multi-modal
inputs
Generic DL
classier; not
CNN-specic
GAN-based
architecture;
not CNN-
focused,
limited to
binary
applications
Six-layer
CNN with
dropout, max
pooling, and
Random Search
optimization,
tailored for
sparse
genomic data
Novelty
Limited by
sequential FS
and classication;
no image
integration
Iterative FS
improves
gene
selection but
lacks multi-
modal data
Combines
metaheuristic
FS (CSSMO)
with mRMR
and DL for
early cancer
detection
Addresses
high-dimensional,
imbalanced
data with
GAN and
feature
bundling
Unies
mRMR,
advanced
image processing,
and tailored
CNN; leverages
multi-modal
data for
enhanced
accuracy and
interpretability
Table 1. Comparative analysis of GONF and recent hybrid approaches for cancer classication using
microarray Data.
Scientic Reports | (2025) 15:39465 4
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
genes sampled from thousands of anatomical sites across six human donor brains, oering a detailed view of
gene activity in specic brain regions. is allows researchers to explore gene regulation, identify spatial patterns
of gene expression, and study the molecular basis of brain functions and disorders. e dataset al.so includes
metadata on the sampled regions, enabling precise localization of gene activity within the brain’s anatomical
structure. In addition to microarray data, the AHBA features structural and functional imaging, such as MRI
scans and histological sections. ese high-resolution images provide a visual context for the gene expression
data, allowing for the integration of molecular proles with brain morphology. is combination supports
multi-modal analyses, such as linking gene expression patterns to brain structure, function, or disease states.
Accessible through the Allen Institute’s portal, the dataset is a powerful resource for researchers studying the
intersection of genomics, neuroscience, and imaging, enabling a deeper understanding of the human brain’s
complexity29.
Proposed method
Within the cancer classication model, the three most important components are feature selection, data
preparation, and classication. Aer the data preprocessing step, we do outlier processing and standardization
on the dataset to make sure all the data are properly formed. To identify important features, a gene selection
procedure based on the mRMR mechanism is used. We create a 4:2 training and test set from the chosen subset,
with the former being fed into our GONF. To be more precise, our deep neural network builds the optimal initial
weights to achieve a better result. Furthermore, the test samples are used to gauge the model’s ecacy. Figure3
clearly illustrates the ow diagram of the proposed method-GONF.
Pre-processing
e purpose of this section is to provide in-depth information regarding the preprocessing that is utilized in
the GONF method to ensure accuracy. Preprocessing is crucial for the eectiveness of the GONF method, as it
directly impacts the reliability of the results obtained. erefore, various techniques are employed to normalize
data and ensure that the input is of superior quality prior to analysis. e image processing methods used in
this study are applied directly to actual histopathological and microarray images. ese techniques play specic,
Fig. 1. Representative examples of e Cancer Genome Atlas (TCGA) glioblastoma frozen sections marked up
for (A, B) necrosis and (C, D) angiogenesis and the distribution of cases within TCGA gene expression classes.
Scientic Reports | (2025) 15:39465 5
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
useful roles in our data pipeline rather than being employed as metaphorical or analogous references. Our
framework makes use of datasets like TCGA and AHBA, which oer high-resolution visual data in addition
to numerical gene expression proles. In particular, AHBA contains anatomical and histological scans of brain
tissue, whereas TCGA provides H&E-stained Whole Slide Images (WSIs). ese image modalities are processed
as part of the computational pipeline and are essential to the analysis. e image data undergoes a number of
preprocessing steps to improve feature extraction and guarantee consistency. In order to ensure correct alignment
before CNN input, the Hough Transform is used to identify and correct image skew in scanned microarray
and histopathological sections. e accuracy of spatial feature localization is increased by detecting microarray
grid patterns and suppressing noise through the use of autocorrelation techniques in conjunction with average
brightness analysis. Furthermore, watershed segmentation allows for more accurate feature extraction by
dening areas of interest within overlapping or intricate visual structures, such as gene spots or cell clusters.
Gene expression data and the resulting preprocessed images are converted into structured representations and
fed into the CNN architecture. is design improves classication performance by allowing joint learning from
gene expression proles and image-derived features.
Normalization
Using z-score normalization, the TCGA and AHBA microarray datasets were preprocessed to standardize gene
expression values. is ensured that each gene had a mean of 0 and a standard deviation of 1 for every sample.
For a gene expression value
xij ,
gene (i)
and sample
(j)
, the z-score was computed as follows:
Fig. 3. Flow diagram of the method.
Fig. 2. Gene expression map of human brain atlas (AHMBA).
Scientic Reports | (2025) 15:39465 6
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
z
ij =
x
ij
−µ
i
σi
where
σi
is the standard deviation and
µi
is the gene
i
mean expression. To lessen the impact of extreme values,
robust outlier clipping was used prior to normalization, capping expression values at the 1st and 99th percentiles.
Pixel intensities were normalized to the range [0, 1] for histopathological and microarray-derived images using
min-max normalization:
I
norm (x, y)=
I(x, y)−I
min
Imax −Imin
the original pixel intensity is represented by
I(x, y)
and the image’s minimum and maximum intensities are
represented by
Imin
and
Imax
or the CNN, images were resized to a uniform input size of
128 ×128 ×1
.
Image preprocessing
To ensure correct alignment, skew was detected and corrected using the Hough Transform (HT) on microarray-
derived visual data and histopathological images. e HT is a mathematical technique used in image processing
for detecting geometric shapes, such as lines, in digital images. It works by transforming points in the image
space into a parameter space, where patterns corresponding to the desired shapes can be identied. is
transformation is particularly eective for detecting lines, as it simplies the problem into nding intersections
or peaks in a two-dimensional parameter space. In the Cartesian coordinate system, a line is expressed as Eq.(1).
y=mx +c
(1)
where m is the slope, and c is the y-intercept. However, this representation is unsuitable for computational
purposes, especially for vertical lines where the slope becomes undened. Instead, the Hough Transform uses
the polar form of a line, given by Eq.(2).
ρ=x·cosθ +y·sinθ
(2)
Here, ρ is the perpendicular distance from the origin to the line, and θ is the angle between the x-axis and the
normal to the line. e coordinates (x, y) are points on the line. is representation ensures uniform handling of
lines, including vertical ones. Also to detect lines, edge points in the image are rst identied using techniques
like the Canny edge detector. Each edge point (x, y) is mapped into the parameter space using the Eq.(3).
ρ(θ)=x·cosθ +y·sinθ
(3)
is results in a sinusoidal curve in the parameter space for each edge point, representing all possible lines
passing through that point. is process is repeated for all edge points, generating multiple curves in the
parameter space.
To identify the lines, the Hough Transform uses an accumulator array, which discretizes the parameter space
into bins. e value of each bin is incremented whenever an edge point contributes to the corresponding (ρ,θ)
pair. is can be represented as Eq.(4).
Accumulator [
i, j
]=∑
δ
(
ρ
−(
xi
·
cosθj
+
yi
·
sinθj
))
(4)
Here, δ is the Dirac delta function, ensuring that only the relevant bins are incremented.
Peaks in the accumulator array correspond to the most prominent lines in the image. Each peak provides
the parameters (ρ,θ) of a detected line. ese parameters can be used to reconstruct the line in the image space
using the Eq.(5).
x·cosθ +y·sinθ =ρ
(5)
e Hough Transform is highly robust to noise and can detect lines even when they are partially obscured. It
is commonly used for tasks such as skew detection in images, where the dominant angles of detected lines help
determine the skew angle of the image. By analyzing these angles, the skew can be corrected by rotating the
image by the negative of the computed angle. is makes the Hough Transform a versatile tool in applications
requiring geometric analysis and alignment.
Calculate average brightness values
In a microarray image, calculating the average brightness values along horizontal or vertical directions involves
computing the mean projection of pixel intensities across one axis. For the horizontal brightness prole, the
average brightness for each row y in an image with dimensions x×y is calculated using the Eq.(6).
B
H
(
y
)=1
x
∑
i
=1
xI
(
i, y
)
(6)
Here, x is the total number of pixels in the horizontal direction, y is the index of the row, and I(i, y) represent the
intensity value of the pixel at column ii in row y. is equation computes the average brightness of all pixels in a
Scientic Reports | (2025) 15:39465 7
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
specic row by summing their intensity values and dividing by the total number of pixels in that row. Similarly,
for the vertical brightness prole, the average brightness for each column x is determined using the Eq.(7).
B
V(x)=1y
∑
j=1yI (x, j
)
(7)
Here, y is the total number of pixels in the vertical direction, x is the index of the column, and
I(x,j)
represent
the intensity value of the pixel at row j in column x. is equation calculates the average brightness of all pixels
in a specic column by summing their intensity values and dividing by the total number of pixels in that column.
Figure 4 shows the process of identifying locations in a microarray block. Figure 4a presents the vertical
histogram of the microarray block based on the mean value of the x pixels, which represents the signal intensity
distribution in the vertical direction. is histogram contains noise that can aect the data analysis. In Fig. 4b,
the autocorrelation function is applied to the histogram, which shows the repetitive pattern present in the data.
Fast image dehazing methods based on linear transformations have improved the sharpness and detail of images
in vision systems. Applying such techniques in the preprocessing of histopathological images can increase the
accuracy of cancer classication30. e peaks of the autocorrelation function provide important information
about the periodic position of the elements in the block.
In Fig. 4c, the noise present in the histogram has been removed using noise removal techniques and a cleaner
pattern of signal intensities can be seen. Finally, in Fig. 4d, grid lines are drawn based on the identied locations
in the data. ese lines represent the designated locations for the microarray elements, which is done to facilitate
more detailed analysis and data alignment. is process plays an important role in increasing the accuracy and
eciency of microarray data analysis.
Gene expression
e gene selection procedure is markedly inecient for accurate classication, but the mRMR method may
signicantly enhance classication accuracy31. In a high-dimensional microarray, the presence of hundreds of
genes renders the direct application of a technique problematic. Furthermore, accurately training a classier
poses considerable diculties. Alternative solutions must be employed to address this situation successfully.
us, mRMR is employed initially to remove noisy and redundant genes. Cross-modal causal learning in
radiology report generation has shown that merging image and text data can improve clinical analysis32. is
approach is similar to our goal of merging genomic and image data to improve cancer classication. is
technique may be utilized for both continuous and discrete datasets to evaluate the signicance and redundancy
of variables and to pinpoint the most advantageous ones. is research oers a comparative examination of
mRMR and the maximum relevance approach (MaxRel), employing several machine learning classiers across
separate microarray datasets. e experimental ndings indicate that mRMR is a highly eective method
for enhancing feature selection ecacy. e features selected by mRMR demonstrate enhanced predictive
potential and produce more precise classication outcomes than those indicated by MaxRel. e integration of
convolution and transformer architectures in image feature extraction signicantly improves the performance
of medical vision models33. Experimental results from comparative cancer microarray datasets demonstrate
that the mRMR lter approach exhibits greater ecacy when employed alongside SVM-RFE. It has been shown
that mRMR can be eectively used with other feature selection techniques, including wrappers. Hybrid models
based on histopathological images and gene mutations have shown accurate performance in predicting survival
of patients with colorectal cancer. ese results are in line with our strategy of integrating imaging and genomic
Fig. 4. Identifying locations within a microarray block. (a) Vertical histogram of a microarray block, (b)
applying autocorrelation function to the histogram, (c) outcomes of noise elimination, (d) outcomes of
gridding.
Scientic Reports | (2025) 15:39465 8
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
data34. is may be implemented to identify a highly condensed subset from prospective characteristics at a
lower expense. Furthermore, the authors in35 introduced a novel gene selection method that combines an mRMR
lter methodology with a genetic wrapper strategy. e ndings of this investigation demonstrated that the
mRMRGA method was superior to both mRMR ltering and GA wrapper through all datasets. Concurrently,
an equivalent number of chosen genes from this experimental outcome demonstrated that the gene set obtained
using mRMRGA selection was more representative of the designated class. We will utilize the mRMR gene
selection method to identify predictive genes that have low redundancy with other genes in the microarray
dataset and maximal relevance to certain cancer classications. e mRMR technique utilized two mutually
benecial data procedures: one to evaluate the signicance between cancer types and individual genes, and the
other to examine the redundancy across gene pairs. Figure 5 illustrates the mRMR dataset, which consists of
the indices of the chosen genes arranged sequentially. e initial row indicates the most relevant and the least
extraneous genes.
e signicance of a selected collection of genes, SG, can be articulated as Eq.(8).
MG
=
1
SG
∑
G
x∈CG
W(Gx,E
)
(8)
W(Gx, F) denotes the value of the shared knowledge between a gene Gx from the SG class and the malignant
cell class E = {e1, e2}, where e1 represents the normal class and e2 signies the malignancy class. Genes can be
selected to exhibit a high degree of reliability, oen referred to as redundancy, among themselves. e redundant
RG of a selected set of genes SG is as Eq.(9).
RG
=
1
SG2
∑
Gx,Gy
∈
CG
W(Gx,G
y
)
(9)
W(Gx,G
y)
represents mutual information across genes x and y, indicating the extent of their interdependence.
e primary objective of employing the mRMR gene selection approach is to discover a specic subset of genes
from SG, denoted as {xi}, that exhibit the highest dependence on the target class or the least dependence on the
selected gene subset SG. Authors in20 propose utilizing the composite objective to identify equitable solutions
for all parties involved. is criterion employs the maximal relevance criterion and the minimal redundancy
criterion as Eq. (10).
max (MR, RE)=MG −RG
(10)
rough this strategy the microarray datasets, which at rst had about 20,000 genes, were made less dimensional.
mRMR chose a subset of 100 genes for the TCGA dataset with 500 samples, producing a feature matrix of size
(500 ×100)
. 80 genes were chosen for the AHBA dataset with 300 samples, resulting in a feature matrix of
size
(300 ×80)
. In order to choose genes with high predictive power and low redundancy, the mRMR process
computed mutual information for relevance (Eq.8) and redundancy (Eq.9). e objective function (Eq.10)
was then optimized. A validation set was used to adjust the number of chosen genes to optimize classication
accuracy while preserving computational eciency.
Segmentation
e backdrop in microarray images pertains to chemical residues present on the chip during a clinical study34.
Inhomogeneous backgrounds provide a signicant challenge in microarray imaging. is work uses the
watershed method to address the problem. is is implemented to integrate the picture with the backdrop and
texture. Recent proteomic studies have shown that chromatin rearrangements can be an eective therapeutic
target in neuroblastoma36. Incorporating this level of omics information could help improve the gene selection
process in cancer classication models. Watershed segmentation is a powerful image processing technique used
to separate and segment regions within an image based on the concept of topographic representation. In this
approach, the intensity values of an image are treated as a surface, where the pixel values represent elevation.
Bright regions are considered peaks or ridges, while dark regions are interpreted as valleys. e algorithm
works by simulating the ooding of water into these valleys. As water lls the basins, barriers are constructed to
prevent the merging of water from dierent basins, eectively segmenting the image into distinct regions. e
representation of the watershed transformation is shown in Fig. 6.
Fig. 5. A mRMR dataset that has the gene number chosen by the mRMR lter method,
Scientic Reports | (2025) 15:39465 9
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
For formulating the transmission, it is assumed that I(x, y) denote the intensity of the pixel at coordinates (x,
y). e goal of watershed segmentation is to identify catchment basins and ridge lines. A catchment basin is a
region where all pixel’s ow towards a common local minimum under the inuence of the gradient of the image.
e gradient magnitude of the image, G(x, y) is used to detect these basins and is given by Eq.(11).
G
(x, y)=
∂I
∂x
2
+
∂I
∂y
2
(11)
Here,
∂I
∂x
and
∂I
∂y
are the partial derivatives of the image intensity in the horizontal and vertical directions,
respectively. e gradient magnitude highlights the edges in the image, where transitions in intensity occur.
Flooding starts at the local minima of G(x, y) and water gradually lls the catchment basins. To prevent the
merging of water from dierent basins, a boundary or “dam” is built at points where the water from adjacent
basins meets. ese boundaries correspond to the watershed lines. e segmentation output can be represented
as a labeled matrix S(x, y) where each label identies a distinct region. As shown in Eq.(12).:
S(x, y)=k, if (x, y)∈K
(12)
where k is a unique identier for each segmented region.
Watershed segmentation is especially eective in separating touching or overlapping objects, such as cells
in microscopy images. However, it is sensitive to noise and over-segmentation, as small intensity variations
can create additional local minima. To address this, preprocessing techniques like smoothing or using markers
(marker-based watershed) are oen applied. Marker-based watershed segmentation uses predened markers
Fig. 6. Representation of the watershed transformation. (a) An image containing three items that cannot be
delineated by a basic threshold. (b) Segmentation of foreground and background. (c) Inverse of image. (d)
Intensity line prole along the illustrated line, demonstrating the lling of the basins to the level where the
yellow and blue regions converge, resulting in the formation of a rst watershed. (e) Watershed transformation.
(f) masked image by using (b).
Scientic Reports | (2025) 15:39465 10
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
to guide the ooding process, with markers being placed at locations of known objects or regions of interest.
Genomic and epigenomic analyses have shown that miRNAs play a key role in tumor heterogeneity and
immune evasion in brain tumors37. is reduces sensitivity to noise and improves segmentation accuracy. e
mathematical formulation for marker-based segmentation incorporates a modied gradient, Gm(x, y), which
enforces the ooding constraints dened by the markers, ensuring that regions align with the markers’ locations.
Classication
In this study, we leverage a deep CNN for the classication phase, specically designed to handle the complexities
of microarray data in cancer detection. CNNs are a type of DNN commonly used for their exceptional ability
to automatically extract hierarchical features from raw data, making them particularly suitable for complex,
high-dimensional datasets like those found in genomics. Typically, a DNN is structured with a feedforward
network, where the backpropagation method plays a crucial role in training the model by adjusting weights to
minimize errors. e TRAPT framework, utilizing deep learning and epigenomic data, has accurately predicted
key transcriptional regulators. Such models oer new avenues for prioritizing genes in cancer classication38.
Over the years, deep neural networks, including CNNs, have been successfully applied across various domains
such as voice recognition, image analysis, and cancer detection, demonstrating their robustness and versatility in
extracting relevant patterns from complex data. What sets CNNs apart is their ability to learn spatial hierarchies
of features, which is particularly useful for microarray data classication. Unlike traditional machine learning
methods, CNNs can automatically discover intricate relationships between genes in microarray data, making
them ideal for handling high-dimensional and sparse datasets. CNNs excel in capturing local dependencies
between genes and identifying relevant features for cancer subtype classication, all while using fewer parameters
compared to other deep learning models. is characteristic enables CNNs to work eciently with structured
genomic data, oering a promising solution for gene selection and cancer detection tasks. Comprehensive pan-
cancer analyses have identied the NTN1 gene as an important factor in the immune and prognostic inuence
of cancers. ese ndings highlight the importance of focusing on key genes in selecting eective traits39.
e CNN architecture employed in this study consists of six convolutional layers, each followed by a Max
Pooling layer to reduce dimensionality and retain essential features. is structure helps the model learn
progressively more abstract representations of the data, essential for accurately distinguishing cancer subtypes.
To prevent overtting and improve generalization, a “dropout” layer is incorporated aer the pooling layers,
randomly deactivating certain neurons during training. Finally, a smoothing layer is applied to generate the nal
feature vector, which serves as the input for the classication phase. is architecture allows for the eective
classication of cancer types based on microarray data, with the exibility to adapt to various genomic datasets.
Figure7 illustrates the CNN design, and Table2 presents the detailed hyperparameters used in the proposed
model. By utilizing deep CNNs for microarray classication, this study aims to enhance both the accuracy and
interpretability of cancer detection models, oering a powerful tool for identifying gene patterns that are crucial
for understanding cancer biology.
is work addresses the prediction of cancer as a binary classication problem. We assess the prediction
accuracy of the model ability using Binary Cross (BIC) as a loss function. Binary classication problems typically
employ BIC. e subsequent Eq.(13) is employed to calculate the loss utilizing BIC.
BIC
=−1
M
M
∑
j
=1
nilog (f(ni)) + (1 −ni) log (1 −b(ni)) (13)
where f(n) is the probability that n, is the binary label. Minimizing loss enhances the likelihood of b(x) for
samples labeled as 1, hence enabling the use of BIC as an indicator of classication quality. Conversely, the
probability of the sample possessing zero labels is diminishing. Minimizing loss is a process that can signicantly
enhance the model’s accuracy. e deep neural network employs the Rectied Linear Unit (ReLU) as activation
function, while utilizing the sigmoid activation function in the output layer. is is performed to guarantee
that the output is conned to the interval [0, 1], which is essential for compatibility with the BIC loss function.
Fig. 7. e structure of deep CNN.
Scientic Reports | (2025) 15:39465 11
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
Moreover, the output is more readily modiable when employing the sigmoid function as opposed to ReLU.
Equations(14) and (15) illustrate the sigmoid and ReLU functions, which are:
ReLu (w)=Max(0,w
(14)
Sig
(w)=
1
1+ e−
w (15)
e pooling layer of a CNN attains invariance and decreases complexity by eliminating duplicate information
by downsampling. e primary approaches employed for pooling are max and average pooling. Average pooling
calculates the mean value of the designated area and utilizes it as the pooling result, whereas max pooling
Layer Filters Kernel Input size Output size Description
Convolutional 1 64 3 × 3 128 × 128 × 1 128 × 128 × 64
Extracts low-
level features
from input
data
Max Pooling 1 2 × 2 128 × 128 × 64 64 × 64 × 64
Reduces
spatial
dimensions,
retains key
features
Convolutional 2 128 3 × 3 64 × 64 × 64 64 × 64 × 128
Further
feature
extraction
with increased
depth
Max Pooling 2 2 × 2 64 × 64 × 128 32 × 32 × 128
Down
sampling to
reduce
dimensionality
Convolutional 3 256 3 × 3 32 × 32 × 128 32 × 32 × 256
Advanced
feature
extraction
from deeper
layers
Max Pooling 3 2 × 2 32 × 32 × 256 16 × 16 × 256 Further
reduction in
spatial size
Convolutional 4 512 3 × 3 16 × 16 × 256 16 × 16 × 512 Higher-level
feature
extraction
Max Pooling 4 2 × 2 16 × 16 × 512 8 × 8 × 512 Continues
reducing
dimensionality
Convolutional 5 512 3 × 3 8 × 8 × 512 8 × 8 × 512
Continues
extracting
features at
a ne level
Max Pooling 5 2 × 2 8 × 8 × 512 4 × 4 × 512
Reduces
spatial
dimensions
for
classication
Convolutional 6 1024 3 × 3 4 × 4 × 512 4 × 4 × 1024
Final feature
extraction
with deeper
network
Max P
ooling 6 2 × 2 4 × 4 × 1024 2 × 2 × 1024
Reduces
dimensions to
nalize
feature
extraction
Dropout 2 × 2 × 1024 2 × 2 × 1024 Regularization
to prevent
overtting
Flatten 2 × 2 × 1024 1 × 4096 Flattens the
feature map
into a 1D vector
Fully
connected 4096 1 × 4096 1 × N
Final
classication
layer (N =
number of
classes)
Table 2. Hyperparameters for deep CNN Classier.
Scientic Reports | (2025) 15:39465 12
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
determines the greatest value within the zone and employs it as the pooling output. is study utilizes the
maximal pooling technique because of its superior capacity to preserve essential information relative to average
pooling. e max pooling method is illustrated in Eq.(16):
O
r
=
M
(
l
1
a,l
2
a,l
3
a,...,l
j
a
)
(16)
Where M is the maximum pooling, l is the element (j) of the pooling area a, and
Or
is the output results
of the pooling. is method facilitates the retention of essential characteristics while diminishing noise and
extraneous information. Following the application of the convolutional layer to the pre-processed input, this
procedure is reiterated a certain number of times based on the network type. Subsequent of these layers, one
or more fully linked layers are employed to do precise mappings of the extracted characteristics. A completely
connected layer is comparable to a convolutional layer; however, it is characterized by a complete link to its
preceding layer, unlike the sparse connections found in convolutional layers. is is comparable to the process
of establishing connections in conventional neural networks. Mendelian randomization studies have conrmed
the causal relationship between genetic markers such as HbA1c and various diseases40. e last layer produces
a 1-dimensional vector, with the number of components in this vector corresponding to the number of
classication categories. e primary role of this type of layer is to do categorization. In this context, the output
of the neural network is employed to compute the network’s loss rate, that subsequently informs the network’s
attributes and facilitates its training. In this process, the network’s output is evaluated against the correct answer
using an error function, and the error rate is computed. e methodology for calculating errors is dened in
Eq. (17):
L(
Q, ˆ
Q
)
=
1
n
n
∑
i=1
l(Zi,f(Xi,N
))
(17)
e cost function
L(
Q,
ˆ
Q
)
measures the penalty paid by erroneously predicting
ˆ
Q
instead of Q. Subsequent to
the computation of the error percentage, the post-propagation procedure commences. At this point, the gradient
of each parameter is calculated utilizing the chain rule, and all variables are adjusted according to their inuence
on the error produced in the network.
Initializing
CNN generally must learn a complex nonlinear model, and dierent initializers oen lead to diverse convergence
rates and results. Adjusting parameters is challenging, as the neural network fails to acquire essential properties
during backpropagation if entire layer weights are set at 0 or 1. Moreover, an excessively large starting value
will lead to an inated gradient, whilst an insucient initial value would produce a disappearing gradient; both
phenomena impair the network’s learning capability35. e aforementioned concerns must be addressed by
choosing a suitable weight initialization method that meets the following criteria.
• Prevent the neuronal activity values of each layer from reaching saturation.
• Prevent the activation values of each layer from approaching zero.
Nonetheless, network optimization may encounter challenges due to the prevalent utilization of the random
normal method for weight initialization. If the random distribution is improperly generated, the output value
of the deep network may converge to zero, resulting in a vanishing gradient. e primary objective is to avert
the convergence of all output values to zero, sustain uniformity in the activation values and gradient variances
throughout each layer during the propagation process, and guarantee that each layer obtains pertinent feedback
during backpropagation. is initialization is useless with the ReLU function. To preserve variance and
guarantee that 50% of the neurons in every layer are activated, we halve the initialization, as referenced in36 and
demonstrated by Eqs. (18) and (19).
w
j∼R
(
−
√6
qj+qj+1
,
√6
qj+qj+1
)
(18)
w
j
∼
R
(
−
√
6
qj+qj+1 ,
√
6
qj+qj+1
)
2
(19)
Where wj denotes the weight of a certain layer. qj represents the quantity of input neurons in layer j. qj+1 denotes
the quantity of output neurons in layer j + 1. e primary distinction between Eqs.(13) and (14) is the division
by 2, which reduces the range of the uniform distribution by a factor of 2. e next section presents a comparison
of many established weight initialization strategies in clear and succinct language. is approach is employed in
our network due to the advantages of the ReLU activation capability.
Optimization of hyperparameters
Hyperparameters refer to the specic conguration choices made during the development of a model’s
architecture. e eectiveness of a model based on neural networks relies on the careful selection of suitable
hyperparameters. Hyperparameter tuning involves the determination of optimal hyperparameters. is research
employs the Random Search heuristic for hyperparameter optimization. is technique determines the ideal
Scientic Reports | (2025) 15:39465 13
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
solution by methodically exploring a hyperparameter search space and conducting experiments with random
parameter combinations. Estimating the position of objects in complex scenes using neural networks shows
that deep models have the ability to resolve ne details in crowded images. is ability could be useful for
improving the interpretation of cellular images in genomic analyses41. e hyperparameters yielding the highest
accuracy values are selected. e model may be trained on optimal parameters free from aliasing, since we
have chosen the hyperparameters using the approach. To provide a fair assessment of the proposed framework’s
performance, we have adjusted the hyperparameters for each comparative method utilized in this study. e
several hyperparameters included in the proposed approach are listed in Table 3.
Evaluation
We utilized the prescribed approach for cancer categorization. e preliminary phase of dataset preparation
entailed data normalization. A feature selection module was subsequently utilized to choose a subset of features,
which was then fed into the DNN for training. To improve the eectiveness and resilience of the model, we
performed tests utilizing several network optimization methodologies.
Evaluation metrics
is section outlines the assessment measures employed to evaluate the eectiveness of our planned research.
In this study, we utilized many known evaluation criteria. e metrics are presented in Eqs. (20) through (28).
Accuracy
=
TP +TN
FN +FP +TN +TP
(20)
P recision
=
TP
TP +FP
(21)
Recall
=
TP
TP +FN
(22)
F
1=
2×Precision ×Recall
Precision +Recall
(23)
FDR =
FP
TP +FP
(24)
FNR
=
FN
TP +FN
(25)
NPV
=
TN
FN +TN
(26)
FPR
=
FP
TN +FP
(27)
MCC
=
(TN ×TP)−(FN ×FP)
√
(TP +FP)(TP +FN)(TN +FP)(TN +FN) (28)
e provided illustrations present key evaluation metrics used in cancer classication. ese metrics are essential
for assessing the performance and reliability of diagnostic models42.
• True Positive (TP): is metric represents the correct identication of cancerous samples as cancer. It indi-
cates that the model successfully detected cases where cancer is present.
Hyperparameter Values
Input size 128 × 128
Number of layers 5
Activation function ReLU
Pooling type MaxPooling
Pooling size (2 × 2)
Dropout rate 0.2, 0.5
Batch size 16, 32, 64
Learning rate 0.001
Number of epochs 100
Loss function MSE
Regularization L2 Regularization
Table 3. Dierent hyperparameters for the proposed method.
Scientic Reports | (2025) 15:39465 14
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
• False Positive (FP): is metric refers to the incorrect classication of non-cancerous samples as cancerous.
It represents instances where the model falsely alarms by diagnosing cancer in healthy individuals, potentially
leading to unnecessary treatments or anxiety.
• True Negative (TN): is metric captures the accurate classication of non-cancerous samples as non-can-
cerous. It reects the model’s ability to correctly identify healthy individuals and avoid misdiagnosis.
• False Negative (FN): is metric signies the incorrect classication of cancerous samples as non-cancerous.
Such errors are critical as they represent missed diagnoses, which could delay necessary treatment and ad-
versely aect patient outcomes.
Results
is part analyzes all evaluation measures presented in part 3.2 and conducts simulations on all datasets outlined
in Sect. 3.1. We do a comparative study utilizing several methodologies, including Generative Adversarial
Networks (GANs), Sparse Independent Component study (SICA), Deep Belief Networks (DBN), Autoencoder-
based Deep Neural Networks (AEDNNs), and Convolutional Neural Networks (CNN), Support Vector Machine
(SVM), Random Forest (RF) and K-Nearest Neighbors (KNN)43–45. is enhances our comprehension of the
ecacy of our technique in comparison to other fundamental options. Additionally, we evaluate the eectiveness
of our proposed method against many recognized algorithms for cancer categorization. Additionally, we do
ablation research to examine the impact of various approaches on the ecacy of our strategy. e Python
programming language has been only utilized for code development. We utilized many Python libraries,
including pandas, NumPy, TensorFlow, and Scikit-learn, for the experimental work. We have furthermore
utilized other public GitHub repositories. An examination was conducted using a PC that had an Intel Core i7
12th generation CPU and 16GB of RAM.
Figures8, 9, 10, 11 and 12 presents an examination of GONF’s success throughout its rst development,
considering various learning rates. e research indicates that the specied strategy outperforms in every
category. e learning rate, models, and performance metrics are modied for the research. e performance
measurements exhibit both positive and negative characteristics. e Net Present Value (NPV) and the Matthews
Correlation Coecient (MCC) are precise performance metrics indicative of favorable outcomes. Contrary
metrics, such as the False Positive Rate (FPR) and the False Negative Rate (FNR), are also examined. e speed
dierential relative to other models has been determined.
As can see in Fig.8, the GONF method has better FPR ratio than other baseline methods. With a FPR of
less than 2%, GONF outperforms CNN by 3%, GAN by 1.5% and SVM, RF by 6%. is metric is essential in
clinical cancer screening systems, as a false positive can lead to psychological distress, unnecessary biopsies,
increased nancial burden, and waste of clinical resources. High FPR also undermines patient trust in diagnostic
systems. Additionally, a high FPR erodes patient condence in diagnostic tools. By using the mRMR gene
selection technique, which eliminates redundant and uninformative features early in the pipeline and lowers
the possibility of incorrect positive classications, GONF is able to achieve its superior FPR. Additionally, the
specially designed CNN architecture of GONF is explicitly trained using a hyperparameter optimization and a
carefully calibrated dropout strategy, increasing its resistance to noise and irrelevant correlations in microarray
data. By concentrating on biologically signicant gene signatures, these architectural decisions enable GONF to
reduce the rate of incorrectly classifying healthy people.
As shown in Fig.9 the GONF also excels in minimizing the FNR compared to other baseline methods. False
negatives mean missing real cancer cases, which can lead to poor prognoses, delayed treatments, and disease
progression, making this metric crucial in cancer diagnostics.
In contrast, GONF reduced FNR by 2% on the TCGA and AHBA datasets, a substantial improvement over
GANs (3.5%), CNNs (5%), SICA (6%), DBNs (4%), AEDNNs (3%) and classical methods (SVM, KNN, RF) by
average of 7%.is signicant reduction is attributed to GONF’s synergistic use of mRMR for robust feature
selection and its optimized CNN architecture for precise classication, making it a more reliable and accurate
framework for cancer classication. e robust hierarchical feature extraction of GONF, which is intended to
Fig. 8. Examining methods according to the FPR parameter.
Scientic Reports | (2025) 15:39465 15
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
identify subtle and intricate gene expression patterns linked to malignancy, is directly responsible for its capacity
to reduce FNR. By combining denoising, image preprocessing, and the watershed segmentation technique, input
quality is greatly improved, increasing the model’s ability to identify subtle but clinically signicant signals.
GONF maintains high sensitivity in detecting true positives, a crucial performance requirement in life-critical
healthcare applications, by making sure that highly informative gene features are maintained through mRMR
and appropriately emphasized during CNN training. Unlike DBNs, which are prone to trapping local minima
during training, or SICA, which lacks robust supervised learning mechanisms, GONF maintains high sensitivity.
GANs, while helpful in data augmentation, oen lack stability in training, leading to increased FNR. e
balanced optimization strategies in GONF, combined with its ability to handle high-dimensional genomic data,
allow for more accurate identication of cancer cases, minimizing false negatives and improving its utility for
clinical diagnostics.
Furthermore, Fig.10 illustrates that the NPV model signicantly outperforms several baseline methods
when applied with an 80% learning percentage. e NPV model outperforms GAN by 20%, CNN by 10%,
Fig. 10. Examining methods according to the NPV parameter.
Fig. 9. Examining methods according to the FNR parameter.
Scientic Reports | (2025) 15:39465 16
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
DBN by 10%, SICA by 25%, AEDNN by 11% KNN by 17 and RF by 14%. ese modications indicate that the
model has improved its negative predictive values, hence enhancing its accuracy and reliability in predicting
negative instances. Robust preprocessing, ecient feature selection, and an optimized deep neural network that
can dierentiate between normal and abnormal gene expressions all work together to make this possible. Even
with complex or noisy input, the model will generalize well thanks to the addition of dropout and regularization,
which further reduce overtting. In clinical screening, where accurately ruling out cancer is just as crucial as
accurately detecting it, the model’s cautious approach to negative classication is particularly helpful. Because
of its high NPV, the GONF method is a reliable tool for early and accurate cancer screening, ensuring that fewer
true cancer cases are missed.
Similarly, Fig.11 illustrates that the FDR approach signicantly outperforms the baseline methods when its
learning percentage is 80%. It surpasses GAN by 17%, CNN by 13%, DBN by 13%, SICA by 22%, and AEDNN by
11%. e FDR technique eectively reduces the false discovery rate, facilitating the model’s capacity for accurate
predictions by minimizing false positives. ese results indicate that the NPV model and FDR outperform
previously utilized prediction approaches.
As shown in Fig.11, GONF achieves a lower FDR compared to other methods. FDR evaluates how many
of the model’s positive predictions are actually incorrect, and thus has direct implications on the precision and
trustworthiness of diagnostic recommendations. A high FDR not only causes patient distress but also diverts
medical attention from genuine cases. GONF’s low FDR is a reection of its comprehensive multi-stage pipeline
that includes denoising, spatial alignment, robust gene ltering, and deep convolutional analysis.
Using mRMR ensures that the selected features are both highly relevant and minimally redundant, eectively
reducing the inclusion of noisy or irrelevant genes that could lead to false discoveries. While GANs and AEDNNs
show FDRs of around 7% and 10% respectively, they lack the ne-tuned feature selection and classication
integration seen in GONF. Traditional CNNs and DBNs exhibit higher FDRs of approximately 8%, with SICA
being the least eective, showing an FDR of 12%. In contrast, GONF consistently achieves an FDR below 14%,
demonstrating its superior precision. GANs, despite their capacity for generating synthetic data, are prone to
instability during training, which increases FDR. SICA’s unsupervised nature and lack of robustness exacerbate
false discoveries. GONF’s balanced framework, capable of precise gene selection and accurate classication,
ensures fewer false positive predictions, cementing its reliability for cancer diagnostics.
MCC is a powerful metric for evaluating classication performance, particularly under imbalanced class
distributions, which are common in cancer datasets. It provides a single value that combines all confusion matrix
Fig. 12. Examining methods according to the MCC parameter.
Fig. 11. Examining methods according to the FDR parameter.
Scientic Reports | (2025) 15:39465 17
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
components, oering a balanced view of model performance. GONF’s superior MCC indicates that it consistently
achieves a strong correlation between true and predicted classications across all categories. is is achieved
through GONF’s layered feature reduction and abstraction, in which early-stage preprocessing uniformity, and
later-stage convolutional layers focus on learning non-linear gene dependencies. Figure12 illustrates that at
the 50% learning stage, the GONF scheme signicantly outperforms several baseline algorithms. GANs and
AEDNNs achieve MCC scores of 0.85 and 0.87, but their lack of robust segmentation and feature renement
limits their ecacy. CNNs and DBNs score slightly lower at 0.83 and 0.82, while SICA performs the worst with
an MCC of 0.78 due to its inability to handle high-dimensional and noisy data eectively. Also, SVM and KNN
with MCC score 0.7 have shown relatively poor performance. In the meantime, the RF algorithm with MCC
score of 0.78 has relatively better performance.
Comparison with other methods
Alongside its comparison to conventional machine learning techniques, we also evaluate the ecacy of GONF
against several novel, state-of-the-art approaches for cancer classication utilizing microarray data46–51. is
approach assesses the assessment criteria and results for both types of data. e responses evaluated in the
AHBA and TCGA datasets are presented adjacent in Figs. 13 and 14. e GONF model is the most precise (Fig.
13) and the second-highest performer in the TCGA dataset for accuracy, recall, and F1 score. is comparison
demonstrates the ecacy of GONF in generating precise predictions for cancer classication tasks, rivaling
other models. One of GONF’s most advantageous attributes is its little computational power need, rendering it
ideal for real-time applications and environments with constrained resources where operational power may be
limited. e model consistently performs favorably across all assessment metrics, including accuracy, precision,
recall, and F1 score, ensuring equitable and reliable performance. Its ecacy in minimizing the number of false
positives and false negatives demonstrates its eectiveness in therapeutic contexts.
Furthermore, the evaluation of the proposed GONF framework on the AHBA dataset highlights its remarkable
eectiveness, as it outperforms all competing methods in both accuracy and precision, as illustrated in Fig.14.
ese results underscore GONF’s strong ability to correctly identify true cancer cases while simultaneously
minimizing the occurrence of false positives. While high accuracy shows the model’s overall robustness in
both cancerous and non-cancerous classications, high precision means fewer patients receive incorrect cancer
diagnoses, reducing needless treatments and the stress they cause. GONF’s multi-stage processing pipeline, which
consists of rigorous gene selection using mRMR, ecient denoising, and advanced preprocessing techniques,
is directly responsible for its high accuracy. e model can eliminate noise or redundancy and concentrate on
the most biologically signicant features thanks to this design, producing predictions that are more reliable and
consistent. GONF exhibits signicant performance in recall and F1 Score, despite not achieving the absolute
highest values in these metrics. When evaluating a diagnostic tool’s comprehensiveness, recall and F1 score are
both essential. e marginally lower recall indicates a chance to improve the detection of all positive instances,
perhaps by using data augmentation techniques or further ne-tuning the feature extraction process. With
Fig. 14. Performance comparison with new methods (AHBA Dataset).
Fig. 13. Performance comparison with new methods (TCGA Dataset).
Scientic Reports | (2025) 15:39465 18
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
only minor compromises in some metrics, GONF’s performance on the AHBA dataset generally validates its
accuracy, dependability, and diagnostic capability. Its steady strength across a number of evaluation criteria
shows that the system is well-balanced and ready for use in genomics-based cancer detection.
Furthermore, Tables 4 and 5 unequivocally demonstrate that GONF yields dependable and ecient
outcomes across several performance metrics. GONF is an eective method for cancer classication because
to its high accuracy and precision, as well as its improvement in memory and F1 Score. e model’s capacity to
deliver precise outcomes across several criteria demonstrates its robustness and utility. It is currently an eective
instrument for clinical decision-making and patient management in cancer categorization utilizing microarray
data.
Statistical analysis
To evaluate the signicance of the reported enhancements, the results obtained from several iterations of
each method were analyzed statistically using IBM SPSS V.26. Additionally, in conjunction with the typical
calculation of descriptive statistics (mean ± SD), MANOVA and Tukey’s test were performed for each evaluation
metric (Accuracy, Precision, Recall, F1, and AUC) to ascertain any signicant dierences in the comparisons
conducted. e selected signicance level was 0.05. e MANOVA test aims to determine whether there is a
signicant dierence among the outcomes. Tukey’s HSD test facilitates the comparison of each pair of means,
allowing us to ascertain whether pairs have a signicant dierence. e results obtained from the MANOVA
analysis are displayed in Table6.
e MANOVA test results inside Table6 show a statistical distinction between the mean values of all tested
metrics which includes Accuracy, Precision, Recall, F1, and AUC. Running the Tukey’s HSD test becomes
necessary when performing additional research about pairwise analysis. HSD test evaluated the importance of
disparities between each pair of algorithms as shown in Table7. e main focus of statistical analysis within this
section evaluates the outcomes from the leading successful strategy against alternative testing methods. Table7
shows the results obtained from the HSD test applied to three pairs consisting of the research strategy and its best
competing approaches. Q represents the studentized range statistic according to the table. e Q score derives
from evaluating the two mean values being studied.
Measure f-ratio p-value
Accuracy 38.0528 < 0.00001
Precision 30.0156 < 0.00001
Recall 441.929 < 0.00001
F1 87.0745 < 0.00001
AUC 160.9321 < 0.00001
Table 6. Results of MANOVA test on obtained values of dierent metrics.
Methods Accuracy Precision Recall F1
46 79 75 85 81
47 77 75 78 78
48 80 79 83 81
49 86 84 85 84
50 83 86 79 83
51 89 91 73 76
22 89 87 90 87
GONF 95 94 92 90
Table 5. Numerical comparison of the new methods with AHBA dataset.
Methods Accuracy Precision Recall F1
46 79 76 84 86
47 85 82 88 87
48 84 82 89 86
49 77 73 79 77
50 83 74 90 83
51 87 86 91 88
22 96 97 90 92
Table 4. Numerical comparison of the new methods with TCGA dataset.
Scientic Reports | (2025) 15:39465 19
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
e results of Tukey’s HSD test reveal a signicant dierence between the superior outcomes achieved by
GONF and those of its equivalents in most instances. However, there were a few instances where the dierence
was not deemed substantial. ese exceptions arose for the AEDNN and CNN algorithms. e Tukey’s HSD test
indicated that the enhancement in precision attained by the suggested method relative to the AEDNN method
is not statistically signicant. A comparable scenario is noted regarding the discrepancies between the outcomes
of the GONF and CNN algorithms, specically with Precision and F1, which were statistically insignicant.
Ablation analyzes
is section conducts ablation research to evaluate the impact of various components of our proposed approach
on the results. We analyze the impacts of segmentation, augmentation, and denoising in proposed methodology.
Eight distinct outcomes were assessed for the subsequent congurations: Proposed GONF (incorporating
segmentation, augmentation, and denoising); GONF excluding denoising, segmentation, and augmentation;
GONF omitting denoising and augmentation; GONF lacking augmentation and segmentation; GONF devoid
of augmentation; GONF absent of denoising; and GONF without denoising. e below algorithms have been
employed for this objective.
• Segmentation: Utilizing Conditional Random Fields (CRF) to identify and segment named items within
data.
• Augmentation: Implementing random insertions, deletions, or character swaps in a data record to produce
new samples.
• Denoising: Utilizing Bayesian reasoning to ascertain the fundamental noise-free data.
e results for the TCGA and AHBA datasets, presented in Tables 8 and 9, respectively, indicate that the
GONF surpasses other model versions. is exceptional result underscores the ecacy of integrating various
methodologies to improve forecast accuracy and dependability. In the scenarios when a single component is
eliminated, GONF without denoising yields the optimal outcomes for the TCGA Dataset, but GONF without
segmentation excels for the AHBA Dataset. is signies that both denoising and segmentation substantially
inuence the model’s performance, but their eects may dier based on the dataset. e removal of two
components indicates that GONF, without both denoising and segmentation, and GONF, devoid of augmentation
and segmentation, attain optimal performance for the TCGA and AHBA datasets, accordingly. is discovery
highlights the necessity of integrating both augmentation and denoising strategies, since their omission reduces
the model’s ecacy in managing complicated data. e combined impacts of segmentation and denoising with
Method b Prec Rec F1
GONF without
Den., Seg.
and Aug. 78 72 87 79
GONF without
Den. and Aug. 79 75 84 80
GONF without
Den. and Seg. 81 78 88 82
GONF without
Aug. and Seg. 83 82 85 83
GONF without
Aug. 85 84 85 85
GONF without
Seg. 90 85 87 91
GONF without
Den. 86 83 91 87
Proposed GONF 96 95 91 94
Table 8. Ablation analyzes for examine the eects of Den (denoising.), seg (segmentation.), and Aug
(augmentation.) for TCGA Dataset.
Pairwise
comparisons
(Method1 : Method2)
Q.05 = 4.242Q.01 = 5.302
Accuracy
dierence Precision
dierence Recall
dierence F1
dierence AUC
dierence
GONF : GAN Q = 16.20
(p = .00000) Q = 7.75
(p = .00005) Q = 41.36
(p = .00000) Q = 20.20
(p = .00000) Q = 26.93
(p = .00000)
GONF : AEDNN Q = 15.20
(p = .00000) Q = 2.55
(p = .37817)* Q = 45.55
(p = .00000) Q = 18.07
(p = .00000) Q = 30. 80
(p = .00000)
GONF : CNN Q = 11.27
(p = .00000) Q = 0.58
(p = .99626)* Q = 1.86
(p =. 00000) Q = 0.89
(p = .98175)* Q = 11.10
(p = .00000)
Table 7. e results of tukey’s HSD test. * Non-signicant dierence.
Scientic Reports | (2025) 15:39465 20
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
augmentation approaches are clear, demonstrating that each element is essential for improving the overall
performance of GONF.
Tables8 and 9 demonstrate that each component uniquely enhances the model’s overall predictive power,
and their collective impact exceeds the aggregate of their separate contributions. e signicance of each
component is evident when examining their interactions and mutual enhancements. Denoising puries the data,
segmentation delineates pertinent features, and augmentation enhances the variability of the training data. e
elimination of any one component undermines this synergy, resulting in a signicant reduction in performance.
is illustrates the eective collaboration of these parts in extracting and enhancing pertinent information
regarding cancer from photos, hence optimizing the model’s predictive accuracy.
Conclusion
is study aimed to develop a comprehensive system, GONF, for cancer classication using DNN and microarray
data, emphasizing enhanced accuracy, eciency, and reliability in clinical applications. GONF integrates
advanced statistical techniques, such as mRMR for feature selection, and deep neural networks to optimize gene
selection and feature extraction processes. e model employs image processing techniques, including Watershed
segmentation, to preprocess microarray data eectively, reducing noise and isolating regions of interest, which
improves the quality of input features. By feeding these rened features into its deep learning architecture,
GONF demonstrated superior performance in cancer classication tasks. e model was tested against state-
of-the-art methods, including GANs, CNNs, DBNs, SICA, and AEDNNs, showing consistently higher accuracy,
recall, and precision. GONF achieved classication accuracy improvements on the TCGA dataset and AHBA
dataset, signicantly outperforming the comparative models. Additionally, it demonstrated lower false positive
and false negative rates, contributing to its robust predictive capability and reliability in clinical diagnostics.
ese results highlight GONF’s potential as a powerful tool for healthcare professionals, improving diagnostic
accuracy, reducing errors, and enabling timely, personalized treatment planning. e model’s ability to process
large datasets eciently further supports decision-making and resource management in healthcare.
While promising, the GONF model faces certain limitations. Its computational complexity may limit
implementation in resource-constrained environments with limited infrastructure or expertise in deep learning.
Additionally, although the model performed well on specic datasets, further validation on diverse and larger
datasets is needed. Future work should focus on extending its applicability by incorporating real-time data
processing, broader datasets, and additional data types such as genomic and lifestyle factors, while rening its
segmentation and preprocessing techniques for greater adaptability.
Data availability
Availability of data and materials: e Cancer Genome Atlas (TCGA) data used in this study are publicly avail-
able through the Genomic Data Commons (GDC) portal: https://portal.gdc.cancer.gov/, accession number:
phs000178.v11.p8.e Allen Human Brain Atlas (AHBA) data are publicly available from the Allen Institute for
Brain Science: https://human.brain-map.org/, donor IDs: H0351.1009, H0351.1015, H0351.1012, H0351.1016,
H0351.2001, H0351.2002.
Received: 11 January 2025; Accepted: 3 October 2025
References
1. Alharbi, F. & Vakanski, A. Machine learning methods for cancer classication using gene expression data: a review. Bioengineering
10 (2), 173 (2023).
Method Acc Prec Rec F1
GONF without
Den., Seg.
and Aug. 71 65 86 74
GONF without
Den. and Aug. 83 81 87 84
GONF without
Den. and Seg. 84 83 86 85
GONF without
Aug. and Seg. 83 83 84 83
GONF without
Aug. 85 86 85 86
GONF without
Seg. 86 82 93 87
GONF without
Den. 86 85 88 87
Proposed GONF 94 95 87 94
Table 9. Ablation analyzes for examine the eects of Den (denoising.), seg (segmentation.), and Aug
(augmentation.) for AHBA Dataset.
Scientic Reports | (2025) 15:39465 21
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
2. Deng, X., Li, M., Deng, S. & Wang, L. Hybrid gene selection approach using XGBoost and multi-objective genetic algorithm for
cancer classication. Med. Biol. Eng. Comput. 60 (3), 663–681 (2022).
3. Khalsan, M. et al. Fuzzy gene selection and cancer classication based on deep learning model (2023). arXiv preprint
arXiv:2305.04883.
4. Mazlan, A. U. et al. A review on recent progress in machine learning and deep learning methods for cancer classication on gene
expression data. Processes 9 (8), 1466 (2021).
5. Bontonou, M. et al. A comparative analysis of gene expression proling by statistical and machine learning approaches. Bioinform.
Adv. (2024).
6. Basavegowda, H. S. & Dagnew, G. Deep learning approach for microarray cancer data classication. CAAI Trans. Intell. Technol. 5
(1), 22–33 (2020).
7. Samiei, M. et al. Classication of skin cancer stages using a AHP fuzzy technique within the context of big data healthcare. J. Cancer
Res. Clin. Oncol. 149 (11), 8743–8757 (2023).
8. Wu, H., Li, X. & Zhang, J. Gene selection and cancer classication using convolutional neural networks. Comput. Biol. Chem. 93,
107487 (2023).
9. Zhang, H., Lin, C. & Wu, F. Cancer classication using deep neural networks and gene selection techniques. Bioinformatics 38 (6),
1814–1822 (2022).
10. Wang, W., Yin, B., Li, L., Li, L. & Liu, H. A low light image enhancement method based on dehazing physical model. Comput.
Model. Eng. Sci. 143 (2), 1595–1616. https://doi.org/10.32604/cmes.2025.063595 (2025).
11. Li, Z., Liu, Q. & Huang, W. A hybrid method combining convolutional neural network and gene selection for cancer classication.
IEEE Access. 9, 109035–109045 (2021).
12. Zhang, R. et al. MvMRL: a multi-view molecular representation learning method for molecular property prediction. Brief.
Bioinform. 25 (4), bbae298. https://doi.org/10.1093/bib/bbae298 (2024).
13. Yao, Y., Gao, J. & Zhang, W. A deep learning approach to cancer diagnosis using gene expression data: a comparative study.
Bioinformatics 38 (7), 1811–1819 (2022).
14. Ghosh, S. & Patel, D. A comparative study of machine learning algorithms for cancer classication using gene expression data. J.
Cancer Res. er. 19 (4), 1078–1086 (2023).
15. Luo, Z. & Li, Z. A hybrid deep learning model for cancer classication using gene selection techniques. Med. Image. Anal. 74,
102350 (2023).
16. Men, X. et al. Exploring the pathogenesis of chronic atrophic gastritis with atherosclerosis via microarray data analysis. Medicine
103 (16), e37798. https://doi.org/10.1097/MD.0000000000037798 (2024).
17. Shah, R. & Patel, P. A novel hybrid deep learning model for gene selection and cancer classication. Comput. Biol. Chem. 98,
107471 (2023).
18. Chen, X. & Jing, R. Video super resolution based on deformable 3D convolutional group fusion. Sci. Rep. 15 (1), 9050. h t t p s : / / d o i .
o r g / 1 0 . 1 0 3 8 / s 4 1 5 9 8 - 0 2 5 - 9 3 7 5 8 - z (2025).
19. Li, Z., Li, F. & Liu, Q. A deep convolutional neural network model for cancer classication using gene expression proles.
Neurocomputing 473, 157–168 (2023).
20. Xu, T., Wu, F. & Zhang, Y. Feature selection and cancer classication using neural networks and microarray data. J. Comput. Biol.
28 (5), 523–533 (2021).
21. Liu, R., Wang, S., Tian, F. & Yi, L. SIR-3DCNN: a framework of multivariate time series classication for lung cancer detection.
IEEE Trans. Instrum. Meas. 74, 2563. https://doi.org/10.1109/TIM.2025.3563000 (2025).
22. Zafar, S., Ahmad, J., Mubeen, Z. & Mumtaz, G. Enhanced lung cancer detection and classication with mRMR-Based hybrid deep
learning model. J. Comput. Biomed. Inf. 7, 2 (2024).
23. Wang, Y. et al. Identication of human microRNA-disease association via low-rank approximation-based link propagation and
multiple kernel learning. Front. Comput. Sci. 18(2), 182903 (2024).
24. Wang, J., Chen, Y. & Zou, Q. Inferring gene regulatory network from single-cell transcriptomes with graph autoencoder model.
PLoS Genet. 19(9), e1010942 (2023).
25. Alsuhimat, F. M. & Mohamad, F. S. A hybrid method of feature extraction for signatures verication using CNN and HOG a multi-
classication approach. IEEE Access. 11, 21873–21882 (2023).
26. Das, A., Neelima, N., Deepa, K. & Özer, T. Gene selection based cancer classication with adaptive optimization using deep
learning architecture. IEEE Access (2024).
27. Zeng, Y., Zhang, Y., Xiao, Z. & Sui, H. A multi-classication deep neural network for cancer type identication from high-
dimension, small-sample and imbalanced gene microarray data. Sci. Rep. 15 (1), 5239 (2025).
28. Zeng, Y. et al. GCCNet: a novel network leveraging gated Cross-Correlation for Multi-View classication. IEEE Trans. Multimedia.
27, 1086–1099. https://doi.org/10.1109/TMM.2024.3521733 (2025).
29. Zhang, Y. & Liu, M. A novel deep neural network model for cancer classication with gene expression data. IEEE Trans. Comput.
Biology Bioinf. 20 (3), 854–863 (2023).
30. Wang, W., Yuan, X., Wu, X. & Liu, Y. Fast image dehazing method based on linear transformation. IEEE Trans. Multimedia. 19 (6),
1142–1155. https://doi.org/10.1109/TMM.2017.2652069 (2017).
31. Li, P. & Yang, Z. A hybrid gene selection approach for cancer classication using deep learning methods. Comput. Biol. Med. 139,
104800 (2022).
32. Chen, W. et al. Cross-Modal causal representation learning for radiology report generation. IEEE Trans. Image Process. 34 ({}),
2970–2985. https://doi.org/10.1109/TIP.2025.3568746 (2025).
33. Yin, L. et al. Convolution-Transformer for image feature extraction. Comput. Model. Eng. Sci. 141 (1), 87–106. h t t p s : / / d o i . o r g / 1 0 .
3 2 6 0 4 / c m e s . 2 0 2 4 . 0 5 1 0 8 3 (2024).
34. He, B. et al. A fusion model to predict the survival of colorectal cancer based on histopathological image and gene mutation. Sci.
Rep. 15 (1), 9677. https://doi.org/10.1038/s41598-025-91420-2 (2025).
35. Zhao, S. & Liu, T. Deep learning-based cancer classication from gene expression proles. Artif. Intell. Med. 114, 101046 (2021).
36. Liu, Z. et al. Proteomic analysis reveals chromatin remodeling as a potential therapeutical target in neuroblastoma. J. Translational
Med. 23 (1), 234. https://doi.org/10.1186/s12967-025-06298-5 (2025).
37. Yang, Z. et al. Integrative analysis of genomic and epigenomic regulation reveals MiRNA mediated tumor heterogeneity and
immune evasion in lower grade glioma. Commun. Biology. 7 (1), 824. https://doi.org/10.1038/s42003-024-06488-9 (2024).
38. Zhang, G. et al. TRAPT: a multi-stage fused deep learning framework for predicting transcriptional regulators based on large-scale
epigenomic data. Nat. Commun. 16 (1), 3611. https://doi.org/10.1038/s41467-025-58921-0 (2025).
39. Luan, F., Cui, Y., Huang, R., Yang, Z. & Qiao, S. Comprehensive pan-cancer analysis reveals NTN1 as an immune inltrate risk
factor and its potential prognostic value in SKCM. Sci. Rep. 15 (1), 3223. https://doi.org/10.1038/s41598-025-85444-x (2025).
40. Han, L. et al. Causal associations between HbA1c and multiple diseases unveiled through a Mendelian randomization phenome-
wide association study in East Asian populations. Medicine 104 (11), e41861. https://doi.org/10.1097/MD.0000000000041861
(2025).
41. Al-Selwi, M. et al. Enhancing object pose Estimation for RGB images in cluttered scenes. Sci. Rep. 15 (1), 8745. h t t p s : / / d o i . o r g / 1 0 .
1 0 3 8 / s 4 1 5 9 8 - 0 2 5 - 9 0 4 8 2 - 6 (2025).
42. Zeng, C. & Zhang, W. Classication of cancer data using deep learning methods: a systematic review. Genetic Eng. Biotechnol. J. 41
(3), 307–319 (2022).
Scientic Reports | (2025) 15:39465 22
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
43. Wang, J. & Li, Y. Hybrid deep learning approaches for gene expression data and cancer classication. Comput. Biol. Chem. 93,
107466 (2021).
44. Wu, P. & Lu, C. Machine learning and deep learning models for cancer classication with microarray gene expression data. J. Clin.
Bioinform. 13 (1), 22 (2023).
45. Zhao, Q. & Sun, Y. A novel deep neural network model for cancer classication using gene expression data. Sci. Rep. 12 (1), 3123
(2022).
46. Zhang, H. & Sun, Z. A hybrid feature selection and classication approach for cancer diagnosis based on gene expression data.
IEEE Trans. Biomed. Eng. 68 (8), 2681–2689 (2021).
47. Li, Y. & Li, P. A convolutional neural network approach for cancer classication based on gene expression. BMC Cancer. 23 (1), 523
(2023).
48. Chen, J. & Zhang, Q. Deep learning for cancer diagnosis using gene expression proles: a review. Artif. Intell. Med. 106, 101778
(2022).
49. Zhang, T. & Wu, Z. Multi-objective genetic algorithm for feature selection in cancer classication using deep learning. Comput.
Biol. Chem. 92, 107412 (2023).
50. Wang, Q. & Guo, Z. A hybrid approach combining feature selection and deep learning for cancer classication. Comput. Biol. Med.
147, 105417 (2024).
51. Tan, Z. & Wu, J. An improved deep learning model for cancer classication using gene expression data. IEEE Access. 11, 4692–4700
(2023).
Acknowledgements
is work was supported by the Natural Science Foundation of Zhejiang Province "Research on key technologies
of digital rehabilitation of Parkinson’s disease tremor symptoms with integrated traditional Chinese and Western
medicine at home" (No. LTGY23H270001); Category A Project of the Medical and Health Science and Technol-
ogy Program of GongShu District, Hangzhou City, Zhejiang Province "Research on the Dynamic Intervention
Model of Self-managed Digital Intelligence for the Elderly" (No.A202407).
Author contributions
All authors contributed to the study conception and design. Data collection, simulation and analysis were per-
formed by " Yuanyuan Zhang 1, Jing Chen 1, Chong Zhang 2. e rst dra of the manuscript was written by
“Yuanyuan Zhang"and all authors commented on previous versions of the manuscript.
Funding
e authors did not receive any nancial support for this study.
Competing interests
e authors declare no competing interests.
Ethical approval
is study used publicly available, de-identied data from two major resources: e Cancer Genome Atlas
(TCGA) and the Allen Human Brain Atlas (AHBA). All data were originally collected with informed consent
and under ethical oversight by the respective data providers. TCGA data were obtained from the National
Cancer Institute and the National Human Genome Research Institute. Only open-access TCGA data were used
in this study, which do not require dbGaP authorization. All data comply with the Health Insurance Portability
and Accountability Act (HIPAA) Privacy Rule and the NIH Genomic Data Sharing Policy. AHBA data were
obtained from the Allen Institute for Brain Science. ese data were collected from postmortem human donors
with appropriate consent and ethical approval by the original investigators. e dataset is fully anonymized and
publicly available. No new human subjects were involved in this study, and no additional institutional ethics
approval was required.
Additional information
Correspondence and requests for materials should be addressed to C.Z.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional aliations.
Open Access is article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives
4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in
any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide
a link to the Creative Commons licence, and indicate if you modied the licensed material. You do not have
permission under this licence to share adapted material derived from this article or parts of it. e images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence
and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to
obtain permission directly from the copyright holder. To view a copy of this licence, visit h t t p : / / c r e a t i v e c o m m o
n s . o r g / l i c e n s e s / b y - n c - n d / 4 . 0 / .
© e Author(s) 2025
Scientic Reports | (2025) 15:39465 23
| https://doi.org/10.1038/s41598-025-23147-z
www.nature.com/scientificreports/
