SIAM: Head and Brain MRI Segmentation from Few High-Quality Templates via Synthetic Training

2026-05-04

AI Summary

Current brain MRI segmentation methods rely heavily on large sets of automatically generated atlases, which introduce systematic biases, struggle to generalize to novel anatomical structures, and lack the capability for fully automatic segmentation of extracranial tissues. To address these limitations, this work proposes the SIAM framework, which leverages only six high-quality manually annotated templates and employs domain randomization in both intensity and shape spaces to synthesize training data. SIAM enables 3D whole-head MRI segmentation of 16 anatomical structures—including cerebrospinal fluid, vasculature, dura mater, skull, and skin—spanning both intracranial and extracranial regions. Notably, SIAM achieves contrast-agnostic whole-head segmentation, matching or exceeding state-of-the-art brain segmentation performance across eight heterogeneous datasets (N=301), while significantly improving cross-contrast consistency and sensitivity to gray matter atrophy.

Abstract

Synthetic training has recently advanced brain MRI segmentation by enabling contrast-agnostic models trained entirely on generated data. However, most existing approaches rely on hundreds of automatically labeled templates, introducing systematic biases and limiting their flexibility to incorporate new anatomical structures. We present the Segment It All Model (SIAM), a 3D whole-head segmentation framework for 16 anatomical structures, trained using only six high-quality, manually annotated templates. SIAM extends domain randomization to both intensity and shape domains: synthetic image generation ensures contrast variability, while high-resolution spatial transformations model anatomical differences in cortical thickness and deep nuclei morphology. Unlike prior synthetic models, SIAM simultaneously segments brain as well as extra-cerebral tissues, including cerebrospinal fluid, vessels, dura mater, skull, and skin, enabling fully automated, preprocessing-free analysis. Evaluation across eight heterogeneous datasets (N=301), that include multiple contrasts (T1-weighted, T2-weighted, CT) and span a wide range of ages, demonstrates that SIAM matches or outperforms state-of-the-art methods for brain structures, in addition to extending automated segmentation to non-brain structures. The model also exhibits superior consistency across contrasts and repeated acquisitions, together with improved sensitivity to subtle gray matter atrophy. We openly release the model and the label templates at https://github.com/romainVala/SIAM.

Problem

• MRI segmentation
• synthetic training
• anatomical structures
• domain randomization
• extra-cerebral tissues

Innovation

• synthetic training
• domain randomization
• few-shot segmentation
• whole-head MRI segmentation
• contrast-agnostic model