SAT3D: an uncertainty-aware volumetric AI foundation model for whole-body tumour segmentation

A lightweight volumetric AI foundation model, SAT3D, trained on 17,075 3D volume-mask pairs, generalized across 11 public tumour segmentation datasets, including out-of-distribution scans; a 3D-Slicer plugin supports interactive use.

• SAT3D combines a shifted-window vision transformer with critic-guided uncertainty-aware training, using confidence maps as dense prompts to guide boundary prediction in ambiguous regions.
• It was benchmarked against vision foundation models, prompt-driven methods, and task-specific approaches across 11 public datasets.
• The model showed robust generalization, including out-of-distribution settings, and is supported by a 3D-Slicer plugin for interactive segmentation.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.