Myopic Maculopathy Analysis Using Multi-task Learning and Pseudo Labeling

Abstract

With the advent of deep learning, research has achieved significant success in various fields of ophthalmology, such as diabetic retinopathy detection, glaucoma detection and vessel segmentation. Despite these advancements, there remains a notable gap in the analysis of myopic maculopathy, which carries severe implications such as potential blindness, mainly due to the scarcity of labeled datasets. In addressing this issue, our work in the Myopic Maculopathy Analysis (MMAC) Challenge 2023 focuses on two key tasks. Task 1, Classification of Myopic Maculopathy, aims to accurately categorize the different stages of myopic maculopathy in fundus images. Task 2, Segmentation of Myopic Maculopathy Plus Lesions, requires precisely delineating the areas affected by myopic maculopathy, providing detailed visual maps of the disease’s manifestation. We leverage multi-task learning and pseudo-labeling techniques to overcome the challenges posed by the limited availability of labeled data. Through effective integration of these methodologies, we achieve 4th place in Task 1 and 3rd place in Task 2, marking significant strides in the automated analysis of myopic maculopathy.