Automated Detection of Myopic Maculopathy in MMAC 2023: Achievements in Classification, Segmentation, and Spherical Equivalent Prediction

TL;DR

MMAC 2023 targets automated detection of myopic maculopathy across three tasks: classification, segmentation, and spherical-equivalent prediction. The authors leverage a contrastive pretraining framework based on SimCLR with a ResNet50 backbone for Task 1, independent lesion-specific segmentation models with TTA for Task 2, and an ensemble of tf_efficientnet backbones with five-fold CV and TTA for Task 3, achieving Top 6, Top 2, and Top 1 respectively. They report robust improvements through pretraining, task-specific backbones, and model ensembling, with evaluation metrics including $QWK$, Macro $F1$, Dice Similarity Coefficient ($DSC$), and $R^{2}$/$MAE$. The work demonstrates clinically relevant performance improvements and provides open-source code for reproducibility and potential integration into ophthalmology workflows.

Abstract

Myopic macular degeneration is the most common complication of myopia and the primary cause of vision loss in individuals with pathological myopia. Early detection and prompt treatment are crucial in preventing vision impairment due to myopic maculopathy. This was the focus of the Myopic Maculopathy Analysis Challenge (MMAC), in which we participated. In task 1, classification of myopic maculopathy, we employed the contrastive learning framework, specifically SimCLR, to enhance classification accuracy by effectively capturing enriched features from unlabeled data. This approach not only improved the intrinsic understanding of the data but also elevated the performance of our classification model. For Task 2 (segmentation of myopic maculopathy plus lesions), we have developed independent segmentation models tailored for different lesion segmentation tasks and implemented a test-time augmentation strategy to further enhance the model's performance. As for Task 3 (prediction of spherical equivalent), we have designed a deep regression model based on the data distribution of the dataset and employed an integration strategy to enhance the model's prediction accuracy. The results we obtained are promising and have allowed us to position ourselves in the Top 6 of the classification task, the Top 2 of the segmentation task, and the Top 1 of the prediction task. The code is available at \url{https://github.com/liyihao76/MMAC_LaTIM_Solution}.

Figures (3)

• Figure 1: Proposed pipeline for Task 1.
• Figure 2: Proposed workflow for Task 3. Gray folds represent the training set (internal), the validation set (internal) by green folds, and the test set (internal) by yellow folds.
• Figure 3: Segmentation performance of MAnet on the validation set of Task 2.