Related Experiment Video
Updated: Nov 24, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.1K
Multiclass retinal disease classification and lesion segmentation in OCT B-scan images using cascaded convolutional
Applied Optics
|December 28, 2020
Summary
This study introduces a novel cascaded convolutional network for joint retinal disease classification and lesion segmentation in optical coherence tomography (OCT) images. The framework improves diagnostic accuracy and lesion detection in ophthalmic computer-aided diagnosis.
Area of Science:
- Ophthalmology
- Medical Imaging
- Computer-Aided Diagnosis
Background:
- Retinal optical coherence tomography (OCT) image analysis is crucial for ophthalmic computer-aided diagnosis.
- Current methods for disease classification and lesion segmentation are performed separately, limiting clinical utility and ignoring feature interdependencies.
Purpose of the Study:
- To develop a unified framework for simultaneously classifying retinal diseases and segmenting lesions in OCT images.
- To improve the accuracy and efficiency of ophthalmic diagnostic tools.
Main Methods:
- Proposed a cascaded convolutional neural network framework.
- Utilized an auxiliary binary classification network to distinguish normal from abnormal retinal images.
- Introduced BDA-Net, a novel U-shaped multi-task network with a bidirectional decoder and self-attention mechanism for analyzing abnormal images.
Main Results:
- Achieved a classification accuracy of 0.9913.
- Demonstrated an approximate 3% improvement in Dice coefficient for lesion segmentation compared to a baseline U-shaped model.
Conclusions:
- The proposed cascaded network effectively integrates disease classification and lesion segmentation for OCT images.
- The novel BDA-Net architecture shows significant potential for enhancing ophthalmic computer-aided diagnosis systems.
