PoCo: A Self-Supervised Approach via Polar Transformation Based Progressive Contrastive Learning for Ophthalmic Disease Diagnosis

TL;DR

A self-supervised method via polar transformation based progressive contrastive learning, called PoCo, for ophthalmic disease diagnosis, which achieves state-of-the-art performance with good generalization ability, validating that the method can reduce annotation efforts and provide reliable diagnosis.

Abstract

Automatic ophthalmic disease diagnosis on fundus images is important in clinical practice. However, due to complex fundus textures and limited annotated data, developing an effective automatic method for this problem is still challenging. In this paper, we present a self-supervised method via polar transformation based progressive contrastive learning, called PoCo, for ophthalmic disease diagnosis. Specifically, we novelly inject the polar transformation into contrastive learning to 1) promote contrastive learning pre-training to be faster and more stable and 2) naturally capture task-free and rotation-related textures, which provides insights into disease recognition on fundus images. Beneficially, simple normal translation-invariant convolution on transformed images can equivalently replace the complex rotation-invariant and sector convolution on raw images. After that, we develop a progressive contrastive learning method to efficiently utilize large unannotated images and a novel progressive hard negative sampling scheme to gradually reduce the negative sample number for efficient training and performance enhancement. Extensive experiments on three public ophthalmic disease datasets show that our PoCo achieves state-of-the-art performance with good generalization ability, validating that our method can reduce annotation efforts and provide reliable diagnosis. Codes are available at \url{https://github.com/wjh892521292/PoCo}.

Figures (5)

• Figure 1: An overview of our PoCo architecture. PoCo inputs $n$ raw fundus images in one mini-batch, and performs random data augmentation twice on each of these images to generate positive pairs. Then polar transformation is applied to the augmented views of the images to generate transformed images. The transformed images are fed to a backbone CNN to extract high-dimensional feature vectors, which are used for the first contrastive loss calculation. Via two FC layers, the feature vectors reduce their dimensions and are used to calculate the second and third contrastive losses with a hard negative sampling strategy. Finally, the low-dimensional features learned by PoCo are used for ophthalmic disease classification by fine-tuning the FC layers.
• Figure 2: Illustrating the pixel mapping from the Cartesian coordinate system (a) to the polar coordinate system (b) by using the polar transformation. (c)-(d) A convolution kernel working on Cartesian coordinates and polar coordinates, respectively.
• Figure 3: Visualization of the spatial distribution of positive and negative samples at different stages by t-SNE. Purple: Positive sample pair. Yellow: Negative sample pairs.
• Figure 4: Visualization of CAM examples of AMD (a) and PM (b) images.
• Figure 5: The contrastive loss learning curves of the PoCo with and without the PoT.