MEMatch: A Semi-supervised Cardiac MRI Segmentation Method Guided by Entropy and Multi-scale Joint Strong-weak Consistency

Abstract

Cardiac MRI is a very powerful technique for the diagnosis of cardiovascular disease. Accurate automated segmentation of cardiac MRI can help improve clinical diagnostic efficiency. Currently, supervised cardiac MRI segmentation methods have achieved brilliant achievements, but they mostly rely on an enormous quantity of labeled samples, which is extremely challenging and expensive to acquire. To alleviate the challenge of manual annotation, semi-supervised based method becomes an effective solution. Although there has been some progress in the semi-supervised segmentation method for cardiac MRI, there still exists a gap in its clinical application, and the accuracy of its segmentation needs further improvement. Based on this fact, we propose a novel semi-supervised cardiac MRI segmentation method MEMatch. MEMatch proposes multi-scale joint strong-weak consistency, which applies strong-weak consistency to the prediction results of multiple scales of the network, to more fully utilize the discrepancy between the outputs of different scales of the same network. Meanwhile, we apply entropy minimization to the average prediction of multiple scales, which enforces the average prediction to generate high-confidence predictions and further reduces the discrepancy among the prediction results from different scales. We evaluate the proposed method on the ACDC and LA datasets and compare it with recent methods. The experimental results on 2D and 3D segmentation demonstrate the effectiveness and superiority of our method in various semi-supervised settings compared to state-of-the-art techniques.