PMT: Progressive Mean Teacher via Exploring Temporal Consistency for Semi-Supervised Medical Image Segmentation

TL;DR

Experimental results on two datasets with different modalities, i.e., CT and MRI, demonstrate that the proposed Progressive Mean Teachers framework outperforms the state-of-the-art medical image segmentation approaches across various dimensions.

Abstract

Semi-supervised learning has emerged as a widely adopted technique in the field of medical image segmentation. The existing works either focuses on the construction of consistency constraints or the generation of pseudo labels to provide high-quality supervisory signals, whose main challenge mainly comes from how to keep the continuous improvement of model capabilities. In this paper, we propose a simple yet effective semi-supervised learning framework, termed Progressive Mean Teachers (PMT), for medical image segmentation, whose goal is to generate high-fidelity pseudo labels by learning robust and diverse features in the training process. Specifically, our PMT employs a standard mean teacher to penalize the consistency of the current state and utilizes two sets of MT architectures for co-training. The two sets of MT architectures are individually updated for prolonged periods to maintain stable model diversity established through performance gaps generated by iteration differences. Additionally, a difference-driven alignment regularizer is employed to expedite the alignment of lagging models with the representation capabilities of leading models. Furthermore, a simple yet effective pseudo-label filtering algorithm is employed for facile evaluation of models and selection of high-fidelity pseudo-labels outputted when models are operating at high performance for co-training purposes. Experimental results on two datasets with different modalities, i.e., CT and MRI, demonstrate that our method outperforms the state-of-the-art medical image segmentation approaches across various dimensions. The code is available at https://github.com/Axi404/PMT.

Figures (4)

• Figure 1: Motivation of our PMT. In particular, the standard MT helps to learn robust features by keeping the consistency between teacher and student networks at the current iteration, while our PMT further helps to learn diverse features by maintaining the difference between student networks at different iterations. As a result, more and more high-fidelity pseudo labels will be generated for semi-supervised medical image segmentation.
• Figure 2: An overview of PMT. We employed a progressive design and utilized the architecture of MT. The PMT framework maintains a data buffer of length $\mathcal{B}$ for cross-temporal training. The total loss function $\mathrm L_\mathrm{total}$ for each network includes supervised losses $\mathrm L_\mathrm{CE}, \mathrm L_\mathrm{DICE}, \mathrm L_\mathrm{aln}$, and unsupervised loss $\mathrm L_\mathrm{U}$, $\mathrm L_\mathrm{T}$.
• Figure 3: Illustration of PLF and DDA. PLF is utilized for comparing the representational capacity of models, while DDA aligns model outputs by examining differences.
• Figure 4: 2D & 3D segmentation visualization of different semi-supervised methods under 10% labeled on LA (upper) and pancreas (bottom) dataset.