Mutual Distillation of Dual-Foundation Models for Semi-Supervised PET/CT Segmentation

2026-06-14 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

MuDuo, a novel mutual distillation framework, significantly advances semi-supervised PET/CT segmentation, a critical task for quantitative analysis and radiotherapy planning in oncology. This framework addresses the high annotation costs associated with medical imaging by leveraging both structural CT and metabolic PET imaging foundation models. Specifically, it employs SAM-Med3D for CT and SegAnyPET for PET, acting as modality-specific generalists. MuDuo distills knowledge from these models into a lightweight student network, effectively bridging the gap between task-specific precision and the segmentation priors of generalist foundation models. The approach eliminates the need for manual prompts and maximizes the utility of unlabeled data, achieving state-of-the-art performance on the AutoPET dataset with only 5 labeled cases.

Key takeaway

For research scientists developing medical image segmentation models, MuDuo offers a compelling strategy to overcome annotation bottlenecks. By integrating dual-foundation models via mutual distillation, you can achieve state-of-the-art performance on complex tasks like PET/CT segmentation using significantly fewer labeled examples. Consider adopting this framework to accelerate model development and deployment in resource-constrained medical imaging environments.

Key insights

Mutual distillation of dual-foundation models improves semi-supervised PET/CT segmentation with minimal labeled data.

Principles

• Foundation models can act as modality-specific generalists.
• Bridging task-specific precision with generalist priors is effective.

Method

MuDuo uses mutual distillation to transfer knowledge from SAM-Med3D (CT) and SegAnyPET (PET) into a lightweight student network, eliminating manual prompts for segmentation.

In practice

• Segmenting organs from PET/CT for oncology.
• Reducing annotation costs in medical imaging.

Topics

• PET/CT Segmentation
• Semi-Supervised Learning
• Foundation Models
• Mutual Distillation
• Medical Imaging
• SAM-Med3D
• SegAnyPET

Code references

• Wu-beining/MuDuo

Best for: Computer Vision Engineer, AI Scientist, Research Scientist

Related on AIssential

• Cross-Source Supervision for Bone Infection Segmentation in Dual-Modality PET-CT — Decoupled dual-source learning with PET-CT fusion improves bone infection segmentation by accommodating diverse clinical annotation philosophies.
• Continual Learning for fMRI-Based Brain Disorder Diagnosis via Functional Connectivity Matrices Generative Replay — A continual learning framework uses generative replay and knowledge distillation to diagnose brain disorders from sequential fMRI data.
• CrossPan: A Comprehensive Benchmark for Cross-Sequence Pancreas MRI Segmentation and Generalization — Cross-sequence domain shift is the primary barrier to robust pancreas MRI segmentation.
• BrainDINO: A Brain MRI Foundation Model for Generalizable Clinical Representation Learning — Large-scale slice-wise self-supervised learning yields a unified, transferable brain MRI representation for diverse neuroimaging tasks.
• A 3D SAM-Based Progressive Prompting Framework for Multi-Task Segmentation of Radiotherapy-induced Normal Tissue Injuries in Limited-Data Settings — A 3D SAM-based framework improves radiotherapy injury segmentation using progressive prompting and a small-target focus loss.
• Dual-Domain Equivariant Generative Adversarial Network for Multimodal CT-PET Synthesis — Dual-domain learning and rotational equivariance significantly enhance multimodal medical image synthesis accuracy and robustness.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.