Check Sheet for MS Thesis Proposal and Topic Approval

1 Title Page
2 Table of contents
3 Problem Statement
4 Aims & Objectives
5 Literature Review
6 Methodology of Research
a) theoretical Studies
b) Experimental Set Up
c) Method of Analysis
d) Result Expected
7 Utilization of Research Results
8 Work Schedule Plan
9 Budget Description
10 References
11 Curriculum Vitae
12 Undertaking
13 Board of Post-Graduate Studies Remarks

Certified that the Synopsis/Topic approval report is according to the items listed in
the check sheet.
Research Scholar: Thesis Supervisor:

Hassan Aslam Dr. Muhammad Javed Iqbal

Signature: Signature:

Date: Date:
 M.Sc. Thesis Proposal

Hard and Soft Exudates Detection from Fundus Images Using Deep
Learning

Submitted By

Hassan Aslam
19-MS-DS-37

Supervised By

Dr. Muhammad Javed Iqbal

Assistant Professor

DEPARTMENT OF COMPUTER SCIENCE

FACULTY OF TELECOMMUNICATION AND INFORMATION
ENGINEERING

UNIVERSITY OF ENGINEERING AND TECHNOLOGY

TAXILA

DECEMBER-2022
Contents
1.Introduction:............................................................................................................................................4
2. Problem Statement:.............................................................................................................................5
3. Literature Review:................................................................................................................................6
4. Aims & Objectives:.............................................................................................................................11
5. Method of Analysis:............................................................................................................................12
6. Utilization of Research Results:........................................................................................................12
7. Proposed Time Schedule:.................................................................................................................12
8.Theoretical Studies:.............................................................................................................................13
9.Experimental Setup:............................................................................................................................13
10.Budget Description:...........................................................................................................................13
11.Conclusions:.......................................................................................................................................13
12. References:.......................................................................................................................................14
13. Curriculum Vitae:..............................................................................................................................16
14.Undertaking:.......................................................................................................................................16
15.Supervisor’s Comments:..................................................................................................................17
16.Board of Post Graduate Studies Remark:......................................................................................17
1.Introduction:

Diabetes Mellitus is a disease in which the ability of human body to produce hormone
insulin is impaired, resulting in elevated levels of glucose in the blood. Diabetes can
cause damage to retina in eye and this damage can produce a disease called diabetic
retinopathy. In this disease, blood vessels in the retina of eye can become swollen and
leakage of eye fluids can occur. The high blood pressure damages the thin vessels and
loss of fluid from these vessels cause bruises in the retinal region. The bruises
produced due to leakage of fluids are classified into exudates, cotton wool spots, micro-
aneurysms and hemorrhages. Diabetic retinopathy is the most common cause of
blindness in diabetic patients and vision loss due to diabetic retinopathy is sometimes
irreversible. More than 80% of the people suffering from diabetes for more than twenty
years are vulnerable to Diabetic Retinopathy. Retinal image processing techniques are
critical in the process of detection and diagnosis of diabetic retinopathy. The detection
of exudates is pivotal to diagnosis of diabetic retinopathy. Recent techniques have
introduced to detect exudates, from fundus images but each technique has some
deficiency in terms of low accuracy, algorithm complexity and limited scope of image
database. The diagnosis of retinopathy at an early stage is necessary for proper
medication so that further complexities and loss of vision can be avoided. If this issue
gets detected at an early stage, the risk of blindness can be reduced by 95 percent.

Keywords:

Diabetic Retinopathy, Diabetes Mellitus, Median Filter, Fundus Image, Optic

Disk, Feature Extraction, Exudates, Neural Networks, DIARETDB1 database,
Histogram Equalization, KNN classifier.
2. Problem Statement:

The Diabetes Mellitus causes various complexities regarding health in human body
such as cardiovascular diseases, Renal Failure and damage to retina in eyes. The
blood vessels in the retina of eye can become swollen and leakage of eye fluids can
occur. A retinal image has three regions that are blood vessels, optic disc (OD) and
optic cup. The blood vessels are spread all over the retina, starting from optic disc. The
high blood pressure damages the thin vessels and loss of fluid from these vessels
cause bruises in the retinal region. Diabetic retinopathy is the most common cause of
blindness in diabetic patients and vision loss due to diabetic retinopathy is sometimes
irreversible. The bruises produced due to leakage of fluids are classified into exudates,
cotton wool spots, micro-aneurysms and hemorrhages. If this issue is diagnosed and
properly treated in early stages, the loss to patient’s retina and vision can be controlled.
The detection of Diabetic Retinopathy by employing efficient image processing
algorithms is important in elimination of risk of blindness in diabetic patients.
Many techniques have introduced to detect exudates, but each technique has some
deficiency in terms of low accuracy, algorithm complexity and limited scope in sense of
image database. An algorithm for detection of diabetic retinopathy should be worked out
which will employ the optimal image processing approaches for various stages of
detection such as pre-processing, image contrast adjustment, image feature extraction,
image classification and post-processing. The proposed method will give optimal
performance in terms of accuracy. The performance evaluation of this approach will be
done on the basis of various performance metrics e.g., Sensitivity, specificity and
accuracy for both local and standard datasets.
3. Literature Review:

There are various algorithms which have been proposed in recent past to address the
problem of detection of diabetic retinopathy. These algorithms can be classified
according to the approaches used at various steps of computer aided diagnosis of
diabetic retinopathy. The statistical image-feature based exudates detection, neural
networks-based image classification and morphological image processing-based
exudates detection algorithms have been proposed recently in many state-of-the-art
implementations. ResNet, VGG, and Efficient-Net pre-trained models which performed
results evaluation using quadratic weighted kappa, which is appropriate for classifying
different stages of diabetic retinopathy based on the severity. It considers the level of
disagreement between actual and predicted labels. It achieved the quadratic weighted
kappa score of 0.85 using the EfficientNet b3 network, surpassing the existing models
like support vector machines, decision trees, convolutional neural networks, and
DenseNet pre-trained model [1]. SEResNeXt32x4d and EfficientNetb3 models are used
for pretraining of the aforementioned neural networks has been done on the ImageNet
dataset. Then, the Diabetic Retinopathy images are migrated to these models. Based
on the dataset already available, the output is ultimately split up into 5 levels according
to the seriousness of the degree of DR. The experimental results show that the training
accuracy of this method can reach as high as 0.91 [2]. Transfer learning are widely
adopted by the researchers to overcome annotated data insufficiency problems and
computational overhead. In this article’s proposed strategy, features are extracted from
fundus images using the pre-trained network VGGNet and combined with the concept of
transfer learning to improve classification performance. To deal with data insufficiency
and unbalancing problems, it employed various data augmentation operations
differently on each grade of DR. The results of the experiment indicate that the
proposed framework (which is evaluated on the benchmark dataset) outperformed
advanced methods in terms of accurateness [3]. Automated detection and screening
offer a unique opportunity to prevent a significant proportion of vision loss in our
population. In recent years, CNNs added into the set of algorithms used to screen for
diabetic disease. CNNs promise to leverage the large amounts of images that have
been amassed for physician interpreted screening and learn from raw pixels. The high
variance and low bias of these models could allow CNNs to diagnose a wider range of
nondiabetic diseases as well [4]. DL models are an abstraction of the human brain
coupled with the eyes. To create a model from scratch and train it is a cumbersome task
requiring a huge number of images. This deficiency of the DL techniques can be
patched up by employing another technique to a task called transfer learning. In this, a
DL model is trained on image metadata, and to learn features it used hundreds of
classes from the DR fundus images. This enables professionals to create models
capable of classifying unseen images into a proper grade or level with acceptable
accuracy. This paper proposed a DL model coupled with different classifiers to classify
the fundus image into its correct class of severity. We have trained the model on IDRD
images and it has proven to show very high accuracy [5]. An algorithm for exudates
detection was presented that employed fuzzy C-means clustering for feature extraction
and eight features were fed into SVM classifier for image classification. The
performance is 97.7% for 89 images of DIARETDB1 dataset [6]. An approach that 6
employed gray level co-occurrence features and gray level histogram moments in
feature extraction stage and support vector machine as classifier provided accuracy of
96%. A set of 18 features was used in this detection method [7]. The fundus images
from Messidor Dataset were subjected to pre-processing. Adaptive median filter was
employed for image de-noising. In feature extraction stage, gray level co-occurrence
matrix (GLCM), gray level run length matrix (GLRLM) and scale invariant feature
transform (SIFT) were extracted. The Adaptive Neuro-Fuzzy Inference System is used
to classify image as affected or non-affected images. ANIFS joints the express data
illustration of a Fuzzy Inference System and the learning power of the artificial neural
network. An optimal threshold process grey wolf optimization (GWO) is used for image
segmentation. This method has attained 92.56% accuracy [8]. The intensity
components of hue, saturation and intensity (HSI) space were extracted from images
and then a median filter was applied for reducing the image noise. The top-hat
morphological filter was used to detect bright parts of image. The exudates detection
was done by using a differentiated Gaussian filter. This implementation has achieved an
accuracy of 98% for a collection of locally developed fundus images [9]. The raw retinal
fundus images of Kaggle Diabetic Retinopathy dataset were processed for green
channel extraction. The image contrast was improved by using histogram equalization.
Fourteen features were extracted from pre-processed images for quantitative analysis.
The exudates area is regarded as most suitable feature for classification of image as
exudates image or non-exudates image [10]. The contrast limited adaptive histogram
equalization was applied on green channel of fundus image for contrast improvement
and image de-noising. The two feature descriptors named as pyramid histogram of
oriented gradients (PHOG) and speeded up robust features (SURF) were used for
feature extraction from fundus images. The canonical correlation analysis was adopted
to merge the features obtained by two feature descriptors. The optimal numbers of
features, to be fed to the classifier, were obtained by particle swarm optimization. The
accuracy is 96.2% for images of a local dataset [11]. The visualization of the retinal
structures is pivotal for efficient detection of lesions present in a retinal fundus image.
The retinal image is processed by high pass filtering, adaptive contrast adjustment and
image enhancement procedures. The exudates detection is difficult due to presence of
blood vessels and optic disk in fundus image so they need to be eliminated. The
exudates have high intensity variations with respect to background and are visible as
bright patterns. The accuracy for e-ophtha EX dataset is 89.4% and 87 % for
DIARETDB1 dataset [12]. The input images from DIARETDB-Version-1 and
MESSIDOR datasets were converted into an HSI image from RGB, and then median
filter was applied to de-noise the image while retaining the quality of the image. The
AHE algorithm is applied improve the image contrast and image the foreground and
background were separated by using Threshold method. The morphological dilation
operation is performed on the image to extract the exudates and finally using GLCM the
features of the exudates are extracted for classification. The GLCM provides optimal
feature extraction procedure [13]. An algorithm for screening of diabetic retinopathy on
the basis of two-field mydriatic fundus photography has been proposed. The optic disc
was detected by multi-level wavelet decomposition and recursive region growing. Blood
vessels are extracted by applying histogram analysis on the two median filtered images.
Red lesions are detected using three stage intensity transformation and white lesions
from multi-level histogram analysis. The observed sensitivity and specificity were
80% and 50% respectively [14]. A dataset of real time retinal images was developed
locally. The retinal images were pre-processed by Contrast Limited Adaptive Histogram
Equalization (CLAHE). The color retinal images were segmented using K-Means
Clustering technique. A set of features based on color and texture was extracted and
image classification was done using support Vector Machine. The correct detection rate
of 96% was achieved [15]. Artificial neural networks (ANNs) are learning algorithms that
are inspired by properties of the biological neural networks. They are a system of
interconnected processing elements called nodes. The artificial neural networks had
shown promising performance in image classification applications. The retinal images
from a local data set of 300 images were color normalized in preprocessing step. The
contrast was improved by employing histogram equalization. The image texture features
were recovered by using Texture Energy Measures. The Cascaded Correlation Neural
Network classifies the image as normal or abnormal by processing of 25 texture
features. It does not use a predefined set of hidden units, but the hidden units get added
up one by one until the error is minimized. The neural network was trained by using
Quick prop algorithm. Accuracy is 92.4% [16]. The main blood vessels were extracted
by a multi-resolution analysis technique based on continuous wavelet transform. The
simple local vessel pattern (SLVP) operator is employed for vessel feature extraction.
The detection of the optic disc center and boundary is effectively done by limiting the
search process to the centerline of the main vessels in fundus image. The extracted
features were fed into a multi-layer perceptron neural network for their classification as
inside-optic disc vessels or outside optic disc vessels. The performance evaluation of
this algorithm is done for images from DRIONS database, DRIVE database, and
DIARETDB1 database. The accuracy of optic disc detection for DRIVE and
DIARETDB1 images is 95% and 97.5% respectively [17]. The morphological operations
depend only on the relative ordering of pixel values in an image. The effective utilization
of morphological image processing operations for detection of diabetic retinopathy has
been reported in various recent implementations. The region of interest was separated
from images of structured analysis of the retina (STARE) and digital retinal image for
vessel extraction (DRIVE) datasets. A gray level morphological opening operation was
applied on grayscale image to remove bright lines in blood vessels. Gabor filter,
Gaussian filter and frangi filter were used to enhance the retinal blood vessels and for
feature extraction. The top-hat filter adjusted the contrast between vessels and
background. The image classification as vessel or non-vessel region is done by K-
means clustering and fuzzy C-means clustering. The post-processing step provides a
binary vessel map. The 95.94% of accuracy on STARE and 95.71% of accuracy on
DRIVE databases is obtained [18]. The fundus images from local dataset were
processed for green channel extraction and Sobel filtering was done for de-noising. The
optic disk and vessel elimination was done by morphological operations. The gray level
co-occurrence matrix was extracted in feature extraction and feed forward neural
network for image classification for mild, moderate and severe disease conditions [19].
The potential candidate lesions of retinopathy were segmented from the retinal image
using a combination of morphological-top-hat and Kirsch edge-detection. In feature
extraction step, 208 features based on texture, color, and intensity information were
obtained. The hybrid simulated annealing was applied to select the optimal feature set
to be used as the input to the ensemble bagging classifier. The performance results of
this approach for a local dataset are an accuracy of 97.08%, sensitivity of 90.90% and
specificity of 98.92% [20]. The presence of diabetic retinopathy lesions leads to
incorrect detection of vessels from retinal images. Retinal blood vessels detection can
be done effectively by morphological component analysis (MCA). The MCA algorithm
was adopted to separate vessels and lesions from each other. The Morlet Wavelet
Transform is applied to enhance the retinal vessels. The final vessel image is obtained
by adaptive threshold operation. The performance for DRIVE and STARE datasets is an
accuracy of 95.23% and 95.90% respectively [21]. The Morphological Component
Analysis (MCA) algorithm was employed to discriminate between normal and
pathological retinal structures. The normal and abnormal retinal images were
distinguished by statistical features of the retinal lesions. The 92.01% sensitivity and
95.45% specificity on the Messidor-1 dataset was obtained [22].

Here is the Comparative analysis of performance based detection of the fundus images
 on the literature for reviewing. The table includes different state of art methods with the
author, dataset used, feature extraction technique, classification method, and
performance in terms of accuracy (%).
Ref Accuracy/
No’s. Authors Dataset Methods Performance
(%)
Akhilesh Kumar
Messidor-1 &APTOS
1 Gangwar & Vadlamani Inception-ResNet-v2 72.33&82.18%
2019
Ravi,2021
Muhammad Kashif
Jabbar, EyePACS
2 VGGNet 96.6%
Ayesha Jabbar &
Zaka Ur Rehman,2022
Diabetic Retinopathy
Gazala Mushtaq &
3 Detection 2015 and DenseNet-169 90%
Farheen Siddiqui,2021
Aptos 2019
Dai, Ling , Wu, Liang Li,
4 IDRD DeepDR 90%
Huating et al.2021
Shengchun Long & e-ophtha EX &
5 FCM & SVM 97%
Xiaoxiao Huang.2019 DIARETDB1
Janaki Sathya D Geetha Fundus camera TRC- BPN, PNN & 90,83.3&86.7
6
K,2019 50DXa SVM %
Godlin Atlas L, Kumar
7 Messidor MRG 92%
Parasuraman,2018
Real images from
S.Antelin Vijila,
8 mammogram screening CCNN 92.4%
2R.S.Rajesh,2018
laboratory
Dan Popescu &
9 DRIVE and DIARETDB1 SLVP 95% & 97.5%
andLoretta Ichim,2017
F. Abdali-Mohammadi & 95.94 &
10
A. Poorshamam 2017
STARE & DRIVE KMC & MC
95.71%

4. Aims & Objectives:

The aim of this research would be to develop a system for the detection of hard and soft
exudates from fundus images using deep learning. Following will be the objectives of
this study:
 To implement an efficient preprocessing techniques on the Messidor-1 diabetic
retinopathy dataset and APTOS 2019 blindness detection.
 To extract optimized feature using transfer learning approach on pre trained
Inception-ResNet-v2.
  To perform the implementation and recognition of fundus images types and the
best suited image for diagnose will be demonstrated.
 To achieve a performance comparison of various algorithms for these stages and
an optimal algorithm will be proposed.
5. Method of Analysis:

Data Collection

Data
Preprocesing

Data
Augmentation

Modelling

Results

6. Utilization of Research Results:

The implementation of optimal diabetic retinopathy detection system will serve as a bio-
medical application. It will aid the diabetic patients for early intimation of various health
complications caused by diabetes mellitus. This will lead to better chances of treatment
and patient can recover from these complications through proper medication. In this
way the outcome of this research will benefit the community.

7. Proposed Time Schedule:

Dec Jan Feb Mar April May

Activity
22 23 23 23 23 23
Collection of Literature
Study of Literature
Analysis of the Proposed
Scheme
Preparation of Model
Implementation of
Model
Analysis & Simulation
Result Formulation
Final Write-up & Thesis
Submission

8.Theoretical Studies:

For the theoretical studies we study the recent research papers and existing techniques for
detection of the hard and soft exudates from fundus images. We also explore the domain of
AI, Machine learning, Deep Learning, and transfer learning.
Our main focus on research papers published in high-ranking journals.

9.Experimental Setup:

Hardware resources consist of RAM 8GB

Software and tools Windows 10 Pro Edition, MATLAB, Python.
Processor Intel Intell(R) core (TM) i5-5200U CPU @ 2.20 GHz
System type 64-bit Operating System, x64-based processor
Dataset Messidor-1, APTOS 2019

10.Budget Description:
The datasets and the materials used in this research are freely available on the internet,
and I will use my system for the research.

11.Conclusions:
According to International Diabetes Federation (IDF), around 425 million people were
suffering from diabetes mellitus around the globe in year 2017. More than 80% of the
people suffering from diabetes for more than twenty years are vulnerable to Diabetic
Retinopathy. The symptoms of diabetic retinopathy are exudates, micro-aneurysms and
hemorrhages. If these symptoms are left untreated, they can lead to loss of vision due
to burst of blood vessels in retina.
12. References:

1. Gangwar, A.K., Ravi, V. (2021). Diabetic Retinopathy Detection Using Transfer

Learning and Deep Learning. In: Bhateja, V., Peng, SL., Satapathy, S.C.,
Zhang, YD. (eds) Evolution in Computational Intelligence. Advances in
Intelligent Systems and Computing, vol 1176. Springer, Singapore.
https://doi.org/10.1007/978-981-15-5788-0_64.
2. S. Ramchandre, B. Patil, S. Pharande, K. Javali and H. Pande, "A Deep
Learning Approach for Diabetic Retinopathy detection using Transfer
Learning," 2020 IEEE International Conference for Innovation in Technology
(INOCON), 2020, pp. 1-5, doi: 10.1109/INOCON50539.2020.9298201.2020.
3. Jabbar MK, Yan J, Xu H, Ur Rehman Z, Jabbar A. Transfer Learning-Based
Model for Diabetic Retinopathy Diagnosis Using Retinal Images. Brain Sci. 2022
Apr 22;12(5):535. doi: 10.3390/brainsci12050535. PMID: 35624922; PMCID:
PMC9139157.2020.
4. Gazala Mushtaq, Farheen Siddiqui “Detection of diabetic retinopathy using deep
learning methodology” International Conference on Recent Innovations in
Engineering and Technology (ICRIET 2020) 4TH-5TH December, Tamil Nadu,
India, 2020.
5. Ling Dai, Liang Wu, Huating Li, Chun Cai, Qiang Wu, Hongyu Kong, Ruhan Liu,
Xiangning Wang, Xuhong Hou, Yuexing Liu, Xiaoxue Long, Yang Wen, Lina Lu,
Yaxin Shen, Yan Chen, Dinggang Shen, Xiaokang Yang, Haidong Zou, Bin
Sheng & Weiping Jia,” A deep learning system for detecting diabetic retinopathy
across the disease spectrum”,2021.
6. Shengchun Long , Xiaoxiao Huang , Zhiqing Chen, Shahina Pardhan , and
Dingchang Zheng , "Automatic Detection of Hard Exudates in Color Retinal
Images Using Dynamic Threshold and SVM Classification: Algorithm
Development and Evaluation" , Hindawi BioMed Research International Volume
2019.
7. Janaki Sathya D1,a, Geetha K , " Enhanced CAE system for detection of
exudates and diagnosis of diabetic retinopathy stages in fundus retinal images
using soft computing techniques" , Polish Journal of Medical Physics and
Engineering, Vol 25, Issue 2, June 2019.
8. Godlin Atlas L and Kumar Parasuraman, “Detection of retinal hemorrhage from
fundus images using ANFIS classifier and MRG segmentation”, Biomedical
Research 2018; 29 (7): 1489-1497.
9. Mahsa Partovi, Seyed Hossein Rasta, Alireza Javadzadeh,” Automatic
detection of retinal exudates in fundus images of diabetic retinopathy
patients”, JARCM, Spring 2016; Vol. 4, No. 2.
10. DILIP SINGH SISODIA, SHRUTI NAIR and POOJA
 KHOBRAGADE,"Diabetic Retinal Fundus Images: Preprocessing and
Feature Extraction For Early Detection of Diabetic Retinopathy", Biomedical
& Pharmacology Journal Vol. 10(2), 615-626 (2017).
11. Joel E. W. Koh et. al. "Automated detection of retinal health using PHOG and
SURF features extracted from fundus images", Applied Intelligence (2018).
12. Jaskirat Kaur,Deepti Mittal,"A generalized method for the segmentation of
exudates from pathological retinal fundus images" , biocybernetics and
biomedical engineering 38( 2018 ) 27 .
13. WEI ZHOU et. Al. "Automatic Detection of Exudates in Digital Color Fundus
Images Using Superpixel Multi-Feature Classification", IEEE Access Volume:
5, 2017.
14. S.Saranya Rubini, Dr.A.Kunthavai,"Diabetic Retinopathy Detection Based on
Eigenvalues of the Hessian Matrix", Procedia Computer Science 47 ( 2015 )
311 – 318.
15. NB Prakash, D Selvathi,"An efficient approach for detecting exudates in
diabetic retinopathy images", Biomedical Research 2016; Special Issue:
S414-S418.
16. S.Antelin Vijila, R.S.Rajesh,"DETECTION OF HARD EXUDATES IN
FUNDUS IMAGES USING CASCADED CORRELATION NEURAL
NETWORK CLASSIFIER", International Journal of Pure and Applied
Mathematics Volume 118 No. 11 2018, 699-706.
17. Dan Popescu, and Loretta Ichim, "Intelligent Image Processing System for
Detection and Segmentation of Regions of Interest in Retinal Images",
Symmetry 2018 Volume 10, issue 3.
18. SHILF. Abdali-Mohammadi and A. Poorshamam, "Automatic Optic Disc
Center and Boundary Detection in Color Fundus Images", Journal of AI and
Data Mining. Vol 6, No1,2018,35-.
19. Carson Lam et. al., "Retinal Lesion Detection with Deep Learning Using
Image Patches", Invest Ophthalmology Vis Sci. 2018.
20. PA JOSHI and P. T. KARULE, "Detection of Hard Exudates Based on
Morphological Feature Extraction", Biomedical & Pharmacology Journal Vol.
11, (2018).
21. B. Sumathy, S. Poornachandr, "Automated DR and prediction of various
related diseases of retinal fundus images", Biomedical Research 2018;
Special Issue: S325- S332.
22. Syna Sreng et. al. "Automated Diabetic Retinopathy Screening System Using
Hybrid Simulated Annealing and Ensemble Bagging Classifier", Applied
Sciences, Vol.8,201.
13. Curriculum Vitae:

Name: Hassan Aslam

Father’s Name: Muhammad Aslam
Registration No: 19-MS-DS-37
Contact: +92 320-5709972
E-mail (s): [email protected]

Courses Passed:

Subject Exam Held in Grade GPA

1 Statistical & Mathematical Methods for 2019 C 2.00
Data Science
2 Machine Learning 2019 C+ 2.33
3 Natural Language Processing 2019 B- 2.7
4 Tools & Technique in Data Science 2020 B+ 3.33
5 Big Data Analytics 2020 B 3.00
6 Advanced Computer Vision 2021 B+ 3.33
7 Data Visualization 2021 B+ 3.33
8 Social Network Analysis 2021 B+ 3.33

14.Undertaking:
I certify that t h e research work titled “Hard and Soft Exudates Detection from Fundus
Images Using Deep Learning” is my work. The work has not, in whole or in part, been
presented elsewhere for assessment. Where material has been used from other
sources it has been properly acknowledged/referred to.

Hassan Aslam

19-MS-DS-37
15.Supervisor’s Comments:

Dr. Muhammad Javed Iqbal

Assistant Professor

16.Board of Post Graduate Studies Remark: