Incremental Cross-Domain Adaptation for Robust Retinopathy Screening via Bayesian Deep Learning

TL;DR

This paper introduces a Bayesian incremental learning framework for retinopathy screening that adapts across domains with few training samples, maintaining prior knowledge and understanding disease relationships.

Contribution

It proposes a novel Bayesian multi-objective incremental adaptation method enabling deep models to learn new retinal pathologies with minimal data while preserving previous knowledge.

Findings

Achieves 0.9826 overall accuracy on six datasets.

Outperforms state-of-the-art methods in retinopathy classification.

Effectively learns 13 retinal pathologies across different scanners.

Abstract

Retinopathy represents a group of retinal diseases that, if not treated timely, can cause severe visual impairments or even blindness. Many researchers have developed autonomous systems to recognize retinopathy via fundus and optical coherence tomography (OCT) imagery. However, most of these frameworks employ conventional transfer learning and fine-tuning approaches, requiring a decent amount of well-annotated training data to produce accurate diagnostic performance. This paper presents a novel incremental cross-domain adaptation instrument that allows any deep classification model to progressively learn abnormal retinal pathologies in OCT and fundus imagery via few-shot training. Furthermore, unlike its competitors, the proposed instrument is driven via a Bayesian multi-objective function that not only enforces the candidate classification network to retain its prior learned knowledge during incremental training but also ensures that the network understands the structural and semantic relationships between previously learned pathologies and newly added disease categories to effectively recognize them at the inference stage. The proposed framework, evaluated on six public datasets acquired with three different scanners to screen thirteen retinal pathologies, outperforms the state-of-the-art competitors by achieving an overall accuracy and F1 score of 0.9826 and 0.9846, respectively.