Morph-SSL: Self-Supervision with Longitudinal Morphing to Predict AMD Progression from OCT

TL;DR

A Deep Learning model is developed to predict the future risk of conversion of an eye from iAMD to nAMD from its current OCT scan, outperforming the same network when trained end-to-end from scratch or pre-trained with popular SSL methods.

Abstract

The lack of reliable biomarkers makes predicting the conversion from intermediate to neovascular age-related macular degeneration (iAMD, nAMD) a challenging task. We develop a Deep Learning (DL) model to predict the future risk of conversion of an eye from iAMD to nAMD from its current OCT scan. Although eye clinics generate vast amounts of longitudinal OCT scans to monitor AMD progression, only a small subset can be manually labeled for supervised DL. To address this issue, we propose Morph-SSL, a novel Self-supervised Learning (SSL) method for longitudinal data. It uses pairs of unlabelled OCT scans from different visits and involves morphing the scan from the previous visit to the next. The Decoder predicts the transformation for morphing and ensures a smooth feature manifold that can generate intermediate scans between visits through linear interpolation. Next, the Morph-SSL trained features are input to a Classifier which is trained in a supervised manner to model the cumulative probability distribution of the time to conversion with a sigmoidal function. Morph-SSL was trained on unlabelled scans of 399 eyes (3570 visits). The Classifier was evaluated with a five-fold cross-validation on 2418 scans from 343 eyes with clinical labels of the conversion date. The Morph-SSL features achieved an AUC of 0.766 in predicting the conversion to nAMD within the next 6 months, outperforming the same network when trained end-to-end from scratch or pre-trained with popular SSL methods. Automated prediction of the future risk of nAMD onset can enable timely treatment and individualized AMD management.

Figures (6)

• Figure 1: (a) An overview of our 2-stage training framework. The motivation behind Morph-SSL is shown in (b) and its details in (c).
• Figure 2: Our Encoder (c) comprises a series of Basic Encoder Blocks (b). Except for the number of output channels in the last layer, both Decoder-D and Decoder-A have the same architecture (e) and consist of a series of Basic Decoder Blocks (d). S3DConv (a) is used as the basic convolution operation in both the Basic Encoder and Decoder Blocks.
• Figure 3: Overview of the TTC Task. (a) CDF of the conversion time $T^*$ can be best modeled using a sigmoidal function. Exact $T^*$ is unknown due to the discrete nature of the visits (red dots) but occurs between the first visit where the eye has converted ($T^+$) and the visit ($T^-$) just before it. (b) The Classifier Network to predict the sigmoidal function parameters.
• Figure 4: Examples of Saliency maps with frozen Morph-SSL weights.
• Figure 5: Qualitative visualization of the linear interpolation between the features extracted from two OCT volumes $\mathbf{I}_t$ and $\mathbf{I}_{t+k}$ of the same eye. A single B-scan from the 3D volume has been depicted for a different eye in each row. The smooth transition in the generated intermediate images demonstrates Morph-SSL's ability to learn semantically meaningful features.