SNR-Adaptive Unified Diffusion for Multi-Task Medical Image Segmentation

2026-07-03

AI Summary

This work addresses the challenges of redundant model training and limited anatomical knowledge sharing in clinical cardiac image analysis across multiple datasets and modalities, where joint training often suffers from semantic label conflicts and gradient imbalance. To overcome these issues, the authors propose UniT-Diff, a unified diffusion-based segmentation framework that employs an 11-channel task-isolated output space to prevent gradient interference. It incorporates a signal-to-noise ratio–adaptive task conditioning (SATC) mechanism to dynamically modulate task guidance strength and introduces a task-type-aware conditional dropout (TTACD) strategy to enable neutral-path routing for domain-generalizable samples. Evaluated on the LA, MMWHS, and MNMS datasets, UniT-Diff achieves Dice score improvements of 0.87%, 1.77%, and 0.88%, respectively, significantly outperforming independently trained baselines while supporting semi-supervised learning, unsupervised domain adaptation, and domain generalization within a single model.

Abstract

Clinical cardiac imaging pipelines currently deploy separate models for each dataset and modality, incurring redundant training costs and precluding knowledge sharing across anatomically related tasks. Consolidating semi-supervised learning, unsupervised domain adaptation, and domain generalisation into one model is therefore a practical necessity, yet naive joint training exposes a fundamental barrier: conflicting label semantics between datasets collapse LA Dice from 90.31\% to 83.38\%, while gradient imbalance across tasks of unequal complexity suppresses the weaker tasks throughout training. We present UniT-Diff, a unified diffusion segmentation framework that resolves these conflicts through three targeted mechanisms. An 11-channel task-specific output space physically partitions label categories, eliminating cross-task gradient sign reversal by construction. SNR-Adaptive Task Conditioning (SATC) scales the task token by the log signal-to-noise ratio of the current diffusion timestep, suppressing domain-specific bias during coarse denoising and restoring full task guidance as the signal clears. Task-Type-Aware Conditional Dropout (TTACD) permanently removes the task token for domain-generalisation inputs, routing them through a shared neutral pathway that draws on cross-dataset cardiac anatomy rather than source-vendor statistics. Under a single parameter set, UniT-Diff surpasses independently trained task-specific baselines on all three benchmarks simultaneously: +0.87\% on LA, +1.77\% on MMWHS, and +0.88\% on MNMS.

Problem

• multi-task segmentation
• label semantics conflict
• gradient imbalance
• domain adaptation
• medical image segmentation

Innovation

• Unified Diffusion
• SNR-Adaptive Task Conditioning
• Task-Type-Aware Conditional Dropout
• Multi-Task Medical Image Segmentation
• Domain Generalisation