Enhanced Medical Image Classification Using LSA and PCA in CNN

Abstract

We are presently living in the era where in medical field, the use of technology plays a major role in disease diagnosis and in treatment. In recent years Medical Image Processing play a significant role in modern diagnostics, where precision and accuracy are of highly important for planning and treatment of diseases. In this study, we present an enhanced approach that integrates Least Squares (LSA) alongside with Principal Component Analysis (PCA) within the Convolutional Neural Network (CNN) framework of deep learning to improve image processing and image resolution for medical diagnostics .Here LSA is employed to reduce the noise to the greater extent and to refine the feature for better clarity, while PCA employed in dimensionality reduction for efficient processing and preserving critical image details and at the same time CNN enables the automatic feature extraction and interpretation of image. Our results demonstrate that this combined LSA and PCA in CNN model offers significant improvement in image processing speed, efficiency in computation, reduction in noise present in the medical image, increasing sharpness of the image for high resolution leads to the accuracy in detection of diseases making it a promising method for advanced and enhanced medical imaging applications.

Full text

Enhanced Medical Image Classification Using
LSA and PCA in CNN
Thasneem Suhaifa S
1
, Faizal Mukthar Hussain S
2
, Karthikeyan R
3*
, Sheik Yousuf T
4
,
Rasina Begum B
5
and Mohammed Uveise S A
6
1
PG Scholar, Department of CSE, Mohamed Sathak Engineering College, Kilakarai, India
2,6
Assistant Professor, Department of CSE, Mohamed Sathak Engineering College, Kilakarai, India
3
Professor, Department of CSE(AIML), Vardhaman College of Engineering, Hyderabad, India
4
Professor, Department of CSE, Mohamed Sathak Engineering College, Kilakarai, India
5
Professor, Department of CSBS, Mohamed Sathak Engineering College, Kilakarai, India
Abstract. We are presently living in the era where in medical field, the use
of technology plays a major role in disease diagnosis and in treatment. In
recent years Medical Image Processing play a significant role in modern
diagnostics, where precision and accuracy are of highly important for
planning and treatment of diseases. In this study, we present an enhanced
approach that integrates Least Squares (LSA) alongside with Principal
Component Analysis (PCA) within the Convolutional Neural Network
(CNN) framework of deep learning to improve image processing and image
resolution for medical diagnostics .Here LSA is employed to reduce the
noise to the greater extent and to refine the feature for better clarity, while
PCA employed in dimensionality reduction for efficient processing and
preserving critical image details and at the same time CNN enables the
automatic feature extraction and interpretation of image. Our results
demonstrate that this combined LSA and PCA in CNN model offers
significant improvement in image processing speed, efficiency in
computation, reduction in noise present in the medical image, increasing
sharpness of the image for high resolution leads to the accuracy in detection
of diseases making it a promising method for advanced and enhanced
medical imaging applications.
1 Introduction
In the field of medicine, advancement in image processing and analysis have revolutionized
the way healthcare professionals diagnose and treat patients. Due to the complexity of
medical images and lack of expert’s variability identification in disease leads to delay in
treatment and treating diseases. Medical imaging, such as histopathological slides, X-rays,
MRI and CT scans provides vital insights into the human body's intricate structures and
functions. However, the full potential of these images is only fully realized and obtained
when effective and efficient image processing and analysis techniques are used.
*
Corresponding author: karthikhonda77@gmail.com
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
© The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative
Commons Attribution License 4.0 (https://creativecommons.org/licenses/by/4.0/).

Deep learning, particularly Convolutional Neural Networks (CNNs), has emerged as a
game-changer by allowing automated feature extraction and image interpretation. This
eliminates the necessity for manual data analysis, which can be a time-consuming and error-
prone process. Deep learning has an ability to learn from vast datasets and do processes
including object recognition, segmentation, and classification. However, while deep learning
is proficient in feature extraction, the processed images often require additional refinement
to ensure diagnostic accuracy. In Medical image noise from data is the major problem we
face now days for acquiring accurate and rapid diagnostic processes[27][28]. Here lies the
significance of the least square approximation technique. This mathematical approach that
can further improve the quality of processed medical images by minimizing noise and
enhancing signal constancy. The another mathematical approach called PCA is used in image
processing for dimensionality reduction and extract only the essential features from the image
data. Especially in the data like medical images. PCA is employed to reconstruct the image
with minimal loss with the help of fewer principal component. Principal Component Analysis
is also used to reduce noise, feature extraction for machine learning, visualization and
preprocessing further analysis. The fusion of methodologies like deep learning and
regularized least square approximation and principal Component Analysis will increase the
precision, speed, and automation in medical image analysis.
By these techniques, this study aims to provide more precise, robust, and efficient tools
for medical professionals, ultimately leading to earlier and more accurate diagnoses,
improved treatment outcomes and better patient care. In the following sections, we will
explore the methodology, experiments, and results that highlight the potential of this
approach in enhancing medical image processing and analysis, with implications for
radiology, pathology, and other medical specialties.
2 Related Work
Hugh Harvey et al [1] describe that CAD with ML Provides better solution for breast cancer
monitoring. By leveraging the consistent high sensitivity and specificity performance of
autonomous systems, in combination with expert human oversight. But it is having the
problem of Accuracy of screening workflow efficiency, and increased recall rates. Ken
Chang et al [2] describes that Medical images are classified using DL and overcome the
obstacles of sharing patient data. But it faces the issues of enhancing the results of the rate of
recurrence of image classification. Wang H et al [3] describes the Comparative analysis of
medical images using PET/CT attributes with diagnostic features. But Diagnostic features,
not analyzed as texture features. Alex Krizhevsky et al [4] describes about the High quality
pictures are analyzed by using deep CNN in the ImageNet LSVRC-2010. Predrag S et al [7]
explain about the restoration of imprecise medical images using Deconvolution and image
storing techniques. But it fails in handling the processes deblurring the image are very
repetitive and time consuming process. It yield poor result on medical image. ZiweiLuo et al
[25] Proposes Deep linear kernel and LR space feature methods for high resolution image
restoration. But it fails to get the better feature and design itself a time taking process.
Comparative analysis between different papers is mentioned and described by the table 1.
3 System Model
3.1 DL Based Health Image Examination
DL firstly collects and integrates the huge amount of data from various resources. Then it
applies pre-processing steps for making it suitable for image analysis and other related
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
2

process. Including DL in medical based image analysis covers certain important points. The
points are moral authorization, information right to use, query the information, information
de-identification, information transport, accuracy manage, ordered information, and
parameters for identifying data. Picture that are processed in medical field is too complicated
than other field pictures. 2D methods are used for most CNN based techniques with around
300X300 pixels.[1]. But definitely we can use larger pixels images than the above-mentioned
pixel value in providing training with respect to CNN But it needs large computational
powers. Training can also be eased by differentiating labels to strong and sickly at dissimilar
levels; for example, measuring of illness from low level affected to high level affected
diseases depends on the environment [2].
Figure 1 describe about the proposed sequential order of applied procedure. Firstly,
received medical images are given to least square algorithm for making the low quality
pictures to high quality pictures and then quality image is then one again smoothed by PCA
by reducing the dimension by selecting the Principal Component for maintaining the crucial
medical details and then it will be given CNN block. The refined image is then passed to
CNN for classification and segmentation purpose. Here CNN perform the first level
interpretation of image by identify the abnormal structure or some other features based on
training set. The second PCA is applied to the post CNN to reduce and redundancies to
extreme level for higher precision image. Then the second LSA is applied to the output image
of Second PCA to further refine the image for interpretation. The final output is obtained
with more clear, sharper and accurate image which is used for diagnostic and treatment
purposes.
Table 1. Literature Study Analysis.
S.
No
Year Author Technique Used Challenges
1 2019 Hugh Harvey et
al[1] CAD with ML
Accuracy of screening workflow
efficiency, and increased recall
rates
2 2018
Ken Chang et al
[2]
DL
Enhancing the results of the rate
of recurrence of image
classification.
3 2017 Wang H et al [3]
Random Forests,
SVM, Adaptive
Boosting, and ANN.
Diagnostic features, not
analyzed as texture features.
4 2012 Alex Krizhevsky
et al [4]
DL based CNN Degradation of network
performance.
5 2018
Agnieszka
Mikołajczyk et al
[5]
ML, CNN and DNN Difficult to handle long quality
image.
6 2016 Varun Gulshan et
al [7] Deep CNN
Difficult to find out the
possibility of applying this
algorithm in medical imaging
7 2015
Predrag S et al
[22]
Deconvolution and
image storing
techniques.
The processes deblurring the
image are very repetitive and
time-consuming process. It
yields poor result on medical
image.
8. 2017
Shimil Joseetal
[23]
least square based
deconvolution,
PSNR and SSIM.
It is not effective in black and
white or less light images and
inefficient in case of 2D or other
multiple dimensional evaluation
technique.
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
3

9 2016
M.Kalpana Devi
et al [24]
Blind
Deconvolution
Algorithm, Lucy-
Richardson
Algorithm, and
CNN.
Hard to speed up the deblurring
process.
10 2022 ZiweiLuo et al
[25]
Deep linear kernel
and LR space
feature
To get the better feature and
design itself a time taking
process.
11. 2017
Murali, Srikanth
& Krishnan, K. &
Vishvanathan,
Sowmya & Kp,
Soman [26]
Regularized least
square method
The row wise and column wise
application of regularized least
square method for medical
denoising.
Fig. 1. Medical Image Analysis Sequence
3.2 CNN based Image Identification
CNN is a class of ML algorithm that are frequently engaged in diverse fields including
medical image identification and analysis. As a part of deep learning, The CNN is a powerful
algorithm and has given new vision for image processing [3]. CNN is one of the techniques
where it can be used by both supervised and unsupervised way of classification [2]. Important
step of CNN is to spot image parameter in three dimensions. Initial two dimensions are
mapped with tallness and breadth of the medical picture and final dimension characterize the
color of the received medical pictures. Hidden layer is sequentially connected with initial
layer of neuron by CNN. Then it will be connected with input array. Kernels are used by
CNN for classification and identification and many umber of kernel are used for the purpose
of getting many features that allow the CNN to provide accurate result. Dimensions of
medical pictures are reduced by pooling concept in CNN. In pooling CNN uses two values
of max value and average value. By using these values CNN reduces the size of overall. The
CNN consists of convolutional and pooling levels in layered structure. Final decision is taken
by CNN by suing fully connected layer with less density soft max layers. At training in CNN
loss is computed by using the comparison between labelled and predicted value.
Segmentation and revere segmentation job of CNN is done by using pooling layer end
point and sampling layers. This is used for reconstructing the medical pictures after end of
total process. Total loss of the model is computed by comparing the reconstructed picture
and labelled picture used at the start. Collecting and organizing data for the CNN process is
not simple one as we are processing the medical images with low and high quality.
Importantly we need labelled data after initial collection process. To handle the above-
mentioned issue data augmentation techniques are used. This actually help us in getting less
number of images for getting the conclusive results by applying the steps of rotate, flip etc.
Definitely CNN is the vital tools against the process of medical image classification and
analysis [4]. Figure 2 describes about the way CNN works with image data. [21][25].
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
4

3.3 Image Classification
Image classification is the one of the major application of CNN because it enables the
computer to automatically categorize and understand the given medical image with the help
of image tagging. This Image tagging help us to find the relevant images and condition in
limited time period. The person who works in the field of radiology facing the biggest
challenge in handling individual patients. And classification is done in many applications for
identifying the diseases. In recent days CNN has an enhanced role in analyzing the medical
pictures that includes vital diseases like cancers. [5]-[6]. DL usually need huge size of data
sample for analyzing as we are processing medical filed pictures we don’t get that much
volume of images. Still we need to get better result in image differentiation and improve the
performance. Perceptron might be better option to work with top level performance for low
volume of data [7]. Figure 3 describes the proposed system architecture where least square
used to maximize the image quality and CNN is used to categorize the medical pictures and
identify the fault.
Fig. 2. Steps Involved in CNN
Fig. 3. Medical Image Classification
3.4 Detecting Object
Process of detecting object is to identify and categorize disease based the data applied at
input. In medical pictures, identification methods are used for pin point the area of the
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
5

affected patient. Object detection integrates categorization and localization for computing
what kind of objects are there with given images and determine the way to handle it. Region
based and regression based method are used by DL in medical image classification.[8]. By
using the selective search algorithm images are transformed to small patches from that
identification process can be initiated for getting the high level result in terms of accuracy.
[9]. Regression can also be used for object detection in ML environment. One of the class of
regression algorithm is YOLO and it is very good at velocity and precision. YOLO uses CNN
to process and analyze the medical image. These procedures are used for detecting and
classifying the medical field at very good rate.[11]. It will be one of the very costly process
to get labelled data and detecting disease from labelled data. To tackle this we need to train
and make the model ready based on high volume of medical pictures from total database and
need to fine-tune the model based on filtered medical images.
3.5 Image Segmentation
Segmenting image is part of image processing and it is having lot applications in the field of
analyzing medical images. This segmentation process is used for differentiating images by
analyzing the individual process. It is exactly help us to identify location where patients get
affected by which disease. Neural network model is used for selection and training
procedures. Then it will be experienced and assessed. Finally complete analysis of the
medical picture is executed. Various DL methods have been proposed for segmenting and
analyzing images with specific parameters. Semantic Segmentation and Instance
Segmentation are the important class in segmenting images in the medical field [11]. Two
kinds of images we may receive for segmentation in medical field. The types of images may
be in form of 2D and 3D [12]. Therefore, 2D and 3DCNN used for segmenting images based
on the input data.
3.6 Least Square Approximation
Medical picture analysis is important in future medical monitoring. While capturing medical
images sometimes it provides less quality pictures with very low pixels that is lead to
equipment issues, improper techniques or because of patient intricate body structures and
complex diseases. Training the staff for obtaining the correct imaging techniques is crucial.
If the medical images are not accurate or efficient it may result in misdiagnosis or incomplete
assessment to patient conditions. To prevent this, image enhancement or image denoise
reduction techniques can be employed to improve the medical image quality.
In medical image processing, the least square method can be used in several ways to solve
various problems, such as image restoration, image denoising, image registration deblurring,
image deconvolution, image compression, image stitching, and feature extraction. It has
many applications in the field of clinical diagnosis. At the time of a picture is classified as
blurring or noise image, the least squares method can be used to estimate unblurred image
with maximum efficiency.
Assume that the following data points are provided:
(x1, y1), (x2, y2), (x3, y3),... (xn, yn) (1)
where all x values are independent variables, and all y values are dependent.
Assume further that d stands for error or deviation from each specified point and that f(x) is
the fitting curve.
Now, we can write:
d1 = y1 − f(x1) (2)
d2 = y2 − f(x2)
d3 = y3 − f(x3)
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
6

-
-
dn = yn – f(xn)
The notion that the sum of squares of all deviations from specified values must be minimum,
i.e., the curve that best fits, is expressed by the least-squares method.
(3)
Let's say the following formula is used first in order to find the equation of line of best fit for
the provided data.
Y = a + bX (4)
is the equation of the least square line.
a's normal equation is:
∑Y = na + b∑X (5)
Standard formula for 'b':
∑XY is equal to a∑X + b∑X2 (6)
We may obtain the necessary trend line equation by solving these two normal equations.
Thus, using the formula
y = ax + b (7)
we may get the line of best fit.
A mathematical model that uses mathematical formulas to represents the degradation
process and the relationship between the observed image and the original image. In the case
of noise, the least squares method can be used to denoise an image. This involves solving an
optimization problem to find an image that best fits the observed noisy image while
minimizing the noise. In terms of image registration or aligning of two or more medical
images, the least squares method can be used to find the optimal transformation (translation,
rotation, scaling) that minimizes the difference between the images. In image compression
techniques, least squares optimization is used to find a representation of the image that
minimizes the distortion. When detecting edges or features in an image, the least squares
method can be applied to fit curves or lines to sets of image pixels. This is frequently used in
PC vision and object identification. This is used to extract the needed features. [13]- [18] The
least squares method can be used to align and blend overlapping image regions to create a
seamless composite image for better understanding.
3.7 Principal Component Analysis (PCA)
Principal Component Analysis (PCA) is a statistical mathematical concept which is used to
simplify the data to the greater extent by reducing its high dimensionality of the data like
images while preserving as much of the original information as much as possible. PCA
achieves this by transforming the original variables or dataset into a new set of uncorrelated
variables or dataset, known as principal components, that are ordered by the amount of
variance they capture from the data.[30]The key concept of PCA includes Dimensionality
reduction where data like images contain high dimensional datasets, where some data points
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
7

are often spread across many variable which causes redundancy in data. PCA is employed
to reduce these dimensions by finding a new principal component for the data points which
varies the most. The principal components are nothing but a new variables which means that
the combination of original variable present in the given data.
PCA reduce the pixel value to few principal component that capture essential feature,
main shape and pattern. Transformation of data into new coordinate are performed using
orthogonal transformation in PCA. The orthogonally is achieved by the decomposition of
eigen value of the covariance matrix where each principal component is associated with in
eigen value and Eigen vector. Here Eigen value denotes the amount of variance and
eigenvector point the direction of the component. Component with higher eigen value has
more variance and the component with larger Eigen values are used for analysis. Thus PCA
is used in image denoising, image compression and image classification purpose.
Mathematical Steps involved in PCA:
1. Standardize the Data: X→standardized XX \to \text{standardized
}XX→standardized X.
2. Compute Covariance Matrix: C=1
n−1
X
T
X
C
= \frac{1}{n-1} X^T X
C
=n−11X
T
X.
3. Compute Eigen values and Eigenvectors: Solve C
vi
=λ
i
v
i
Cv_i = \lambda_iv_iC
vi
=λ
i
v
i
for eigen values λ
i
\lambda_iλi and eigenvectors viv_ivi.
4. Select Top kkk Eigenvectors: Choose eigen vectors associated with the largest kkk
eigen values.
5. Project Data: Xreduced=XVkX_{\text{reduced}} = X V_kXreduced=XVk.
3.8 Hybridization of least Squares (LS),Principal Component (PCA) and CNN
The combined methods of LS-CNN can be used to restore the high resolution of the input
data using LS and then passed on to the convolutional layer to identify the medical conditions.
This method showcases an idea of how this method works as follow. Here we start with LS
or a traditional computer vision technique to preprocess and extract relevant features or
enhance the input data from low resolution or blurred image to high resolution or deblurred
image. LS methods are often used for data fitting or feature extraction in image processing.
After the LS preprocessing, you extract important features from the data.
The output from LS is fed as the input to the PCA. Here PCA perform the dimensionality
reduction and the image is reconstructed with minimal loss with losing its original features.
These features are fed as input to a CNN. Feature extraction is a crucial step, as it helps
reduce the dimensionality of the input image and capture relevant information. The take out
attributes are then provided to CNN module, which is model based on DL model designed
for tasks like image classification, segmentation, or object detection. The hybrid model is
trained on a labelled dataset using a combination of LS, PCA and CNN techniques. You
might use LS for initial fitting, PCA for feature extraction and image reconstruction and CNN
for further refinement through back propagation. This trained hybrid model can be used for
performing tasks like image classification, object detection, or segmentation. This hybrid
approach can leverage the strengths of both LS, PCA and CNNs, combining traditional signal
processing with deep learning capabilities.
3.9 LS-PCA-CNN Algorithm
Figure 4 represents the flow of the overall algorithm that has been used for this research.
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
8

3.9.1 Least Squares
 Step 1: Initialization of independent and dependent variables - x
i
, y
i
 Step 2: Computing average of x
i
and y
i
.
 Step 3: Presume the equation of the line of best fit as y = mx + c, m -slope
and c - intercepts.
 Step 4:mis computed by using the below formula:
m = [Σ (X – x
i
) ×(Y – y
i
)] / Σ(X – x
i
)
2
 Step 5: c is computed using the below formula:
c = Y – mX
Fig. 4. Algorithm Flowchart
3.9.2 PCA
 Step 1: Standardize the Data
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
9

 Step 2: Compute Covariance Matrix
 Step 3: Compute Eigenvalues and Eigenvector
 Step 4: Select Top kkk Eigenvectors
 Step 5: Project Data
3.9.3 CNN
 Step 6: Selection of Dataset
 Step 7: Dataset Training
 Step 8: Creation of Trained Data
 Step 9: Dataset Shuffling
 Step 10: Fixing Labels
 Step 11: Normalization of X
 Step 12: Split X and Y for Use in CNN
 Step 13: Training CNN Model
 Step 14: Correctness of Trained Model
4 Results and Discussion
Figure 5 and figure 6 describe about how the image is analyzed and detected the problems in
X-Ray based images and also represent about image classification based on CNN algorithm.
Fig. 5. CNN Based Classification
Fig. 6. CNN Based Detection
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
10

5 Conclusion and Future Enhancement
This article has reviewed many researches related to the medical image detection and
classification using CNN and other machine learning approach. Already these methods
provide some level of accuracy in classifying the medical images. By combining the
algorithm concept of least square approximation, principal component analysis and
convolutional neural network we can further enhance the detection and accuracy level in
medical imaging field. In future we can apply the smart hybrid algorithm for medical image
classification to improve the accuracy to the further level. In the future direction idea
challenging part will be in handling computational cost.
References
1. Harvey, Hugh, Andreas Heindl, Galvin Khara, Dimitrios Korkinof, Michael O’Neill,
Joseph Yearsley, Edith Karpati, Tobias Rijken, Peter Kecskemethy, and Gabor Forrai.
"Deep learning in breast cancer screening." Artificial intelligence in medical imaging:
opportunities, applications and risks (2019): 187-215.
2. Chang, Ken &Balachandar, Niranjan& Lam, Carson & Yi, Darvin& Brown, James &
Beers, Andrew & Rosen, Bruce & Rubin, Daniel &Kalpathy-Cramer, Jayashree.
(2018). Distributed deep learning networks among institutions for medical imaging.
Journal of the American Medical Informatics Association: JAMIA. 25.
10.1093/jamia/ocy017
3. Wang, H., Zhou, Z., Li, Y. et al. Comparison of machine learning methods for
classifying mediastinal lymph node metastasis of non-small cell lung cancer from 18F-
FDG PET/CT images. EJNMMI Res 7, 11 (2017). https://doi.org/10.1186/s13550-017-
0260-9
4. Krizhevsky, A., I. Sutskever, and G. Hinton. 2012. “ImageNet Classification with
Deep Convolutional Neural Networks.” Neural Information Processing Systems 25 (1):
84–90. https://doi.org/10.1145/3065386
5. A. Mikołajczyk and M. Grochowski, "Data augmentation for improving deep learning
in image classification problem," 2018 International Interdisciplinary PhD Workshop
(IIPhDW), Świnouście, Poland, 2018, pp. 117-122, doi:
10.1109/IIPHDW.2018.8388338.
6. Lakhani P, Sundaram B. Deep Learning at Chest Radiography: Automated
Classification of Pulmonary Tuberculosis by Using Convolutional Neural Networks.
Radiology. 2017 Aug;284(2):574-582. DOI: 10.1148/radiol.2017162326. PMID:
28436741.
7. Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, Venugopalan
S, Widner K, Madams T, Cuadros J, Kim R, Raman R, Nelson PC, Mega JL, Webster
DR. Development and Validation of a Deep Learning Algorithm for Detection of
Diabetic Retinopathy in Retinal Fundus Photographs. JAMA. 2016 Dec
13;316(22):2402-2410. doi: 10.1001/jama.2016.17216. PMID: 27898976.
8. Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM, Thrun S. Dermatologist-
level classification of skin cancer with deep neural networks. Nature. 2017 Feb
2;542(7639):115-118. doi: 10.1038/nature21056. Epub 2017 Jan 25. Erratum in:
Nature. 2017 Jun 28;546(7660):686. PMID: 28117445; PMCID: PMC8382232.
9. Yun J, Park JE, Lee H, Ham S, Kim N, Kim HS. Radiomic features and multilayer
perceptron network classifier: a robust MRI classification strategy for distinguishing
glioblastoma from primary central nervous system lymphoma. Sci Rep. 2019 Apr
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
11

5;9(1):5746. doi: 10.1038/s41598-019-42276-w. PMID: 30952930; PMCID:
PMC6451024.
10. S. Ren, K. He, R. Girshick and J. Sun, "Faster R-CNN: Towards Real-Time Object
Detection with Region Proposal Networks," in IEEE Transactions on Pattern Analysis
and Machine Intelligence, vol. 39, no. 6, pp. 1137-1149, 1 June 2017, doi:
10.1109/TPAMI.2016.2577031
11. J. Redmon, S. Divvala, R. Girshick and A. Farhadi, "You Only Look Once: Unified,
Real-Time Object Detection," 2016 IEEE Conference on Computer Vision and Pattern
Recognition (CVPR), Las Vegas, NV, USA, 2016, pp. 779-788, doi:
10.1109/CVPR.2016.91.
12. J. Redmon and A. Farhadi, "YOLO9000: Better, Faster, Stronger," 2017 IEEE
Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA,
2017, pp. 6517-6525, doi: 10.1109/CVPR.2017.690.
13. R. Girshick, J. Donahue, T. Darrell and J. Malik, "Rich Feature Hierarchies for
Accurate Object Detection and Semantic Segmentation," 2014 IEEE Conference on
Computer Vision and Pattern Recognition, Columbus, OH, USA, 2014, pp. 580-587,
doi: 10.1109/CVPR.2014.81.
14. Yang S, Sun Y, Chen Y, Jiao L (2012) Structural similarity regularized, and sparse
coding based super-resolution for medical images. Biomed Signal Process Control
7(6):579–590
15. A. Bovik, The Essential Guide to the Image Processing, Academic Press, New York,
NY, USA, 2009.
16. Honggang Chen, Xiaohai He, Linbo Qing, QizhiTeng, Chao Ren, SGCRSR:
Sequential gradient constrained regression for single image super-resolution, Signal
Processing: Image Communication, Volume 66,2018,Pages 1-18,ISSN 0923-5965.
https://doi.org/10.1016/j.image.2018.04.012.
17. S. Esedoglu and S. J. Osher, “Decomposition of images by the anisotropic Rudin-
Osher-Fatemi model,” Communications on Pure and Applied Mathematics, vol. 57, no.
12, pp. 1609–1626, 2004.
18. M. R. Banham and A. K. Katsaggelos, "Digital image restoration," in IEEE Signal
Processing Magazine, vol. 14, no. 2, pp. 24-41, March 1997, doi: 10.1109/79.581363.
19. R. C. Gonzalez and R. E. Woods, Digital Image Processing Prentice Hall, 3rd edition,
2007
20. Puttagunta, M., Ravi, S. Medical image analysis based on deep learning approach.
Multimed Tools Appl 80, 24365–24398 (2021). https://doi.org/10.1007/s11042-021-
10707-4
21. Pawan Kumar Mall, Pradeep Kumar Singh, SwapnitaSrivastav, Vipul Narayan, Marcin
Paprzycki, Tatiana Jaworska, Maria Ganzha, A comprehensive review of deep neural
networks for medical image processing: Recent developments and future opportunities,
Healthcare Analytics, Volume 4, 2023, 100216, ISSN 2772-4425,
https://doi.org/10.1016/j.health.2023.100216.
22. Predrag S. Stanimirović, Igor Stojanović, Vasilios N. Katsikis, Dimitrios Pappas,
ZoranZdravev, "Application of the Least Squares Solutions in Image Deblurring",
Mathematical Problems in Engineering, vol. 2015, Article ID 298689, 18 pages, 2015.
https://doi.org/10.1155/2015/298689
23. S. Jose, N. Mohan, V. Sowmya and K. P. Soman, "Least square based image
deblurring," 2017 International Conference on Advances in Computing,
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
12

Communications and Informatics (ICACCI), Udupi, India, 2017, pp. 1453-1457, doi:
10.1109/ICACCI.2017.8126045.
24. Devi, M. Kalpana and R. Ashwini. “An Analysis on Implementation of Various
Deblurring Techniques in Image Processing.” (2016).
25. S. F. M. Hussain, R. Karthikeyan, S. Ramamoorthi, I. S. Arafat and S. S. M. Gani,
"Denial of Service Attack Analysis Using Machine Learning Techniques," 2023 2nd
International Conference on Edge Computing and Applications (ICECAA), Namakkal,
India, 2023, pp. 886-896, doi: 10.1109/ICECAA58104.2023.10212215.
26. Z. Luo, H. Huang, L. Yu, Y. Li, H. Fan and S. Liu, "Deep Constrained Least Squares
for Blind Image Super-Resolution," 2022 IEEE/CVF Conference on Computer Vision
and Pattern Recognition (CVPR), New Orleans, LA, USA, 2022, pp. 17621-17631,
doi: 10.1109/CVPR52688.2022.01712.
27. R. Karthikeyan, B. Sundaravadivazhagan, Robin Cyriac, Praveen Kumar
Balachandran, S. Shitharth, "Preserving Resource Handiness and Exigency-Based
Migration Algorithm (PRH-EM) for Energy Efficient Federated Cloud Management
Systems", Mobile Information Systems, vol. 2023, Article ID 7754765, 11 pages,
2023. https://doi.org/10.1155/2023/7754765.
28. Venkatesan, Karunakaran, Pramod Kumar Gouda, Bibhuti Bhusan Rath, and
Murugaperunal Krishnamoorthy. "Optimal day-ahead scheduling of microgrid
equipped with electric vehicle and distributed energy resources: SFO-CSGNN
approach." Journal of Energy Storage 102 (2024): 113933.
29. Medical image denoising and classification based on machine learning: A review,
Aumya Chaturvedi1, Aditya Sai Srinivas T2, Karthikeyan R3, Vijayaraj M3, Nirmal
Kumar A4 and Sangeetha M5,2022 ECS - The Electrochemical Society, ECS
Transactions, Volume 107, Number 1,DOI 10.1149/10701.6111ecst.
30. Murali, Srikanth& Krishnan, K. & Vishvanathan, Sowmya & Kp, Soman. (2017).
Image denoising based on weighted regularized least square method. 1-5.
10.1109/ICCPCT.2017.8074388.
31. Mishra, Sidharth & Sarkar, Uttam & Taraphder, Subhash & Datta, Sanjoy& Swain,
Devi & Saikhom, Reshma& Panda, Sasmita & Laishram, Menalsh. (2017). Principal
Component Analysis. International Journal of Livestock Research. 1.
10.5455/ijlr.20170415115235.
32. Howley, Tom & Madden, Michael & O'Connell, Marie-Louise & Ryder, Alan. (2005).
The Effect of Principal Component Analysis on Machine Learning Accuracy with
High Dimensional Spectral Data. 209-222. 10.1007/1-84628-224-1_16.
E3S Web of Conferences 619, 02013 (2025)
ICSGET 2025
https://doi.org/10.1051/e3sconf/202561902013
13