This research focuses on developing an automated system to classify healthy and macular degeneration fundus images, and localize the affected regions. The key highlights are:
The dataset consists of 598 fundus images, including 299 macular degeneration and 299 healthy fundus images. Data augmentation techniques like horizontal flip, vertical flip, and random rotation were applied to increase the training data.
Seven deep learning models were benchmarked - a standalone Convolutional Neural Network (CNN) model, and six CNN models with different ResNet architectures (ResNet50, ResNet50v2, ResNet101, ResNet101v2, ResNet152, ResNet152v2) as the backbone.
The performance of the models was evaluated using metrics like accuracy, precision, sensitivity, and F1-score. The CNN model with ResNet50 backbone achieved the highest training accuracy of 98.7% for a 90% train and 10% test data split.
Grad-CAM visualization was applied on the best performing model (ResNet50+CNN) to localize the affected regions in the macular degeneration fundus images. This helps medical practitioners easily identify the diseased areas.
The study demonstrates the effectiveness of deep learning techniques in automating the classification and localization of macular degeneration, which can aid early detection and treatment of the disease.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Tahmim Hossa... at arxiv.org 04-25-2024
https://arxiv.org/pdf/2404.15918.pdfDeeper Inquiries