Distilling Temporal Coherence into 2D Networks for Transrectal Ultrasound Prostate Video Segmentation

2026-06-30

AI Summary

This work addresses the trade-off between temporal inconsistency in 2D approaches and high inference latency in 3D methods for prostate segmentation in transrectal ultrasound (TRUS) videos. The authors propose a temporal consistency learning framework that injects temporal context into a 2D network via knowledge distillation during training, preserving efficient single-frame inference while enhancing temporal stability. Key innovations include an optical flow–guided confidence-weighted temporal consistency loss, a contrastive learning–driven dual-scale prototype alignment module, and a geometric-equivariant pseudo-labeling strategy to reduce reliance on densely annotated data. Experiments demonstrate that the method achieves state-of-the-art segmentation accuracy and temporal consistency on both the SUN-SEG benchmark and the newly introduced TRUS-V dataset (2,679 frames), while meeting real-time inference requirements.

Abstract

Real-time video segmentation of the prostate in Transrectal Ultrasound (TRUS) is essential for image-guided interventions. While conventional 2D methods suffer from inter-frame inconsistencies by disregarding temporal context, 3D architectures incur prohibitive latency. To resolve this dilemma, we present a Temporally Consistent Learning Framework that distills temporal coherence into a 2D network during training, preserving single-frame inference efficiency. Our design is driven by a key clinical observation: the prostate exhibits geometric stability, whereas the surrounding acoustic environment fluctuates due to physiological motion and transducer pressure. Because conventional temporal constraints propagate erroneous gradients from these unstable regions, we introduce a Confidence-Weighted Temporal Consistency objective derived from optical flow warping residuals, selectively attenuating contributions from unreliable regions. Complementing this pixel-wise constraint, a Dual-scale Prototype Alignment Module enforces semantic coherence through contrastive optimization of local boundary and global semantic features. Furthermore, to eliminate the need for dense per-frame video annotations, we employ geometric equivariance-based pseudo-labeling with knowledge distillation from a pretrained teacher. Extensive experiments on SUN-SEG and our newly introduced TRUS-V benchmark (2,679 frames) demonstrate state-of-the-art accuracy and temporal consistency at real-time speed. Code and dataset are available at https://github.com/DYDevelop/DTC-TRUS.

Problem

• temporal consistency
• prostate segmentation
• transrectal ultrasound
• real-time video segmentation
• inter-frame inconsistency

Innovation

• Temporal Consistency
• Knowledge Distillation
• Confidence-Weighted Loss
• Dual-scale Prototype Alignment
• Pseudo-labeling