DualFete: Revisiting Teacher-Student Interactions from a Feedback Perspective for Semi-supervised Medical Image Segmentation

TL;DR

This work introduces a feedback mechanism into the teacher-student framework to counteract error reconfirmations, and proposes a dual-teacher feedback model, which allows more dynamics in the feedback loop and fosters more gains by resolving disagreements through cross-teacher supervision while avoiding consistent errors.

Abstract

The teacher-student paradigm has emerged as a canonical framework in semi-supervised learning. When applied to medical image segmentation, the paradigm faces challenges due to inherent image ambiguities, making it particularly vulnerable to erroneous supervision. Crucially, the student's iterative reconfirmation of these errors leads to self-reinforcing bias. While some studies attempt to mitigate this bias, they often rely on external modifications to the conventional teacher-student framework, overlooking its intrinsic potential for error correction. In response, this work introduces a feedback mechanism into the teacher-student framework to counteract error reconfirmations. Here, the student provides feedback on the changes induced by the teacher's pseudo-labels, enabling the teacher to refine these labels accordingly. We specify that this interaction hinges on two key components: the feedback attributor, which designates pseudo-labels triggering the student's update, and the feedback receiver, which determines where to apply this feedback. Building on this, a dual-teacher feedback model is further proposed, which allows more dynamics in the feedback loop and fosters more gains by resolving disagreements through cross-teacher supervision while avoiding consistent errors. Comprehensive evaluations on three medical image benchmarks demonstrate the method's effectiveness in addressing error propagation in semi-supervised medical image segmentation.

Figures (7)

• Figure 1: Pre-experiments on the LA dataset. 16 labeled samples are used. FullySpervise runs with only the labeled set, while others use both. Results are evaluated on the first unlabeled sample and averaged from the last 100 training steps. It can be seen that medical image segmentation is susceptible to continual errors, so the confirmation bias issue is fairly problematic. Consistent errors can be reduced by the feedback interaction [highlighted in (d)].
• Figure 2: Schematic of the feedback mechanism. (a) Feedback is applied to each unit's likelihood, leading to a uniform updating direction. (b) The dual-teacher framework enables the feedback more dynamic based on prediction confidence.
• Figure 3: Visualizations of several methods (10% labels).
• Figure 4: Disagreement between two teachers (disag) and the pseudo-label error (PL) [Eq. \ref{['eq:dualt:pl']}], measured by 1-Dice. We report results evaluated by training inputs (first row) and by the testing set (last row), respectively. (LA, 10% labels).
• Figure 5: Effects of confidence thresholding (in $\mathcal{L}^{\mathcal{A}}_{cs}$) and strong-augmentation likelihood (in $\mathcal{L}_{df}$).