Page 36 - AIH-1-4

Artificial Intelligence in Health

ORIGINAL RESEARCH ARTICLE
Deep learning-powered segmentation and

classification of diabetic retinopathy for
enhanced diagnostic precision

Manoj Saligrama Harisha , Arya Arun Bhosale* , and M. Narender

Department of Computer Science and Engineering, The National Institute of Engineering, Mysuru,
Karnataka, India

Abstract

This study addresses the critical challenge of diabetic retinopathy (DR), a severe
complication of diabetes that potentially leads to blindness. We introduce a novel
approach to DR detection using transfer learning, leveraging a single fundus
photograph to automatically identify the disease’s stage. DR progresses through four
stages, posing challenges for early detection, with existing methods often inefficient
and prone to disagreements among clinicians. The proposed approach demonstrated
in the APTOS 2019 Blindness Detection Competition employs convolutional neural
networks (CNNs) and achieved a high quadratic weighted kappa score of 0.92546,
highlighting its effectiveness in automatic DR detection and emphasizing the need for
timely intervention. This paper first reviews related work, spanning classical computer
vision methods to deep learning approaches, with a focus on CNNs. Transfer learning
*Corresponding author: with CNN architectures is explored, showcasing promising results from various studies.
Arya Arun Bhosale
(2020cs_aryaarunbhosale_a@nie. Identifying two critical gaps in existing literature, the research emphasizes the need
ac.in) for comprehensive exploration into integrating pre-trained large language models
(LLMs) with segmented image inputs for generating test/treatment recommendations.
Citation: Harisha MS, Bhosale AA,
Narender M. Deep learning- In addition, understanding the dynamic interactions among integrated components,
powered segmentation and including lesion segmentations, disease classification, and LLMs within web applications,
classification of diabetic retinopathy remains essential. The objectives of the study include developing a comprehensive DR
for enhanced diagnostic precision.
Artif Intell Health. 2024;1(4):30-42. detection methodology, exploring and implementing model integration, evaluating
doi:10.36922/aih.2783 performance through competition ranking, contributing significantly to DR detection
Received: January 19, 2024 methodologies, and identifying research gaps. The study encompasses revolutionizing
DR detection by integrating cutting-edge technologies, focusing on transfer learning
Accepted: April 1, 2024
and various model integrations within web applications. The methodology covers data
Published Online: September 6, pre-processing, augmentation, segmentation using U-Net neural network architecture,
2024 and a detailed training process. The U-Net model demonstrates efficient segmentation
Copyright: © 2024 Author(s). of retinal structures with high accuracy and an impressive frames-per-second rate. The
This is an Open-Access article results highlight the model’s effectiveness in segmenting blood vessels, hard exudates,
distributed under the terms of the
Creative Commons Attribution soft exudates, hemorrhages, microaneurysms, and the optical disc, with high Jaccard, F1,
License, permitting distribution, recall, precision, and accuracy scores. These findings underscore the model’s potential
and reproduction in any medium, to enhance diagnostic capabilities in retinal pathology assessment, promising improved
provided the original work is
properly cited. patient outcomes through timely diagnosis and intervention in combating DR.
Publisher’s Note: AccScience
Publishing remains neutral with Keywords: Diabetic retinopathy; Deep learning; Segmentation; Transfer learning;
regard to jurisdictional claims in
published maps and institutional Convolutional neural networks; Lesion segmentations; Disease classification; U-Net architecture
affiliations.

Volume 1 Issue 4 (2024) 30 doi:10.36922/aih.2783