diabetic-retinopathy-severity-xai

KO | EN

gitlite — search

#ai #typescript #llm #golang #go #android #deep-learning #linux #docker #cli #kubernetes #vue #openai #open-source

diabetic-retinopathy-severity-xai

(★ 14)

Explainable deep learning approach for Diabetic Retinopathy severity classification from retinal fundus images.

diabetic-retinopathy-severity-xai 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

EXPLAINABLE AI FOR DIABETIC RETINOPATHY SEVERITY DETECTION

Overview

Diabetic Retinopathy is a major visual complication in diabetic patients, which can lead to irreversible loss of vision if not treated properly. This project aimed at developing an automatic system capable of analyzing fundus images of the retina, grading the severity of diabetic retinopathy, and providing visual explanations to justify the model decisions.

Apart from prediction accuracy, the aim of this work is interpretability, an important factor in clinical and medical decision-support systems.

Problem Statement

Manual screening of the retina is a time-consuming process and requires ophthalmologists who have undergone training. This has resulted in delays in diagnosis, especially in large-scale programs.
This project covers the use of deep learning for:

Grading the severity of Diabetic Retinopathy

Highlighting the important retinal regions that influence the prediction

Dataset

The experiments are conducted on retinal fundus images sourced from the APTOS 2019 Blindness Detection dataset, which categorizes images into five severity levels:

->No DR

->Mild DR

->Moderate DR

->Severe DR

->Proliferative DR

Basic preprocessing and data organization were performed to ensure consistency and reliability during training and evaluation.

Methodology

The system follows a structured pipeline:

1.Data Exploration & Preprocessing

->Inspection of class distribution

->Image normalization and resizing

->Handling class imbalance using weighted loss

2.Model Training

->A convolutional neural network architecture was used for feature extraction and classification

->Multi-class classification was performed across five DR severity levels

3.Evaluation

->Performance evaluated using accuracy and macro F1-score

->Validation used to monitor generalization behavior

4.Explainability

->Visual explanation techniques (Grad-CAM) were applied

->Heatmaps highlight retinal regions contributing to the model’s decision

Results

The model demonstrates reasonable performance in distinguishing between different DR severity levels.
More importantly, the explainability component provides transparent visual insights, helping to bridge the gap between automated predictions and clinical trust.

Disclaimer

This project is intended for educational and research purposes only and is not a replacement for professional medical diagnosis.

Author

Muhammed Sinan

diabetic-retinopathy-severity-xai

diabetic-retinopathy-severity-xai 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

diabetic-retinopathy-severity-xai

diabetic-retinopathy-severity-xai 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README