BrainGFM improves brain image tasks with generative AI model

A generative foundation model for computational brain imaging

Artificial Intelligence

Summary

Brain imaging techniques like MRI and CT scans often face challenges because there isn’t enough consistent patient data for many tasks. The authors designed BrainGFM, a new AI model trained on over a million brain scan slices that can generate realistic and useful brain images. This model helps improve the accuracy of brain scan analysis and image quality, especially when there isn’t much data to work with. In tests involving thousands of stroke patients, BrainGFM's generated images were nearly as good as real scans and helped doctors classify stroke types better.

What this means in practice

  • For hospital radiology teams: Produce high-quality synthetic brain images and improve stroke subtype classification where clinical brain imaging data are limited.
  • For medical imaging device developers: Enhance reconstruction algorithms in MRI and CT scanners using a pretrained generative model adaptable to diverse image types and qualities.
  • For neurotechnology startups: Integrate generative priors for multi-task brain imaging applications to support diagnostic tools and AI-human collaboration in clinical settings.$Commercial implications: Enables development of advanced diagnostic software marketed to hospitals enabling improved clinical decision support leveraging synthetic imaging.

Authors

Yuting He, Xu Wen, Hongxing Lan, Chengliang Liu, Yuxing Liu, Yan Xi, Wenxue Yu, Jindi Kong, Zhiqiang Zhang, Liang Jiang, Rongjun Ge, Yang Chen, Multi-Center Cooperative Clinical Study Group for "Lesion-Phenotype Brain Mapping in Acute Stroke"

Abstract

Computational brain imaging is constrained by a mismatch between increasingly diverse tasks and scarce, heterogeneous clinical data. Although existing brain foundation models transfer pretrained representations, their knowledge remains entangled within model-specific features and fixed architectures. This paper proposes BrainGFM, a generative foundation model that extends knowledge transfer beyond representation reuse through a reusable and reconfigurable brain-imaging prior. By pretraining on a corpus of about 1.3 million MRI and CT slices, BrainGFM supported task-aligned data synthesis for low-data predictive learning and task-adapted post-training for image reconstruction. Across 48 downstream tasks, BrainGFM improved segmentation and classification under limited supervision and achieved high-fidelity image enhancement across heterogeneous imaging conditions. Multicenter clinical validation involving 4,796 stroke patients across two internal and four external centers demonstrated its applicability to three clinically relevant tasks around stroke cohorts. Blinded reader assessments showed that BrainGFM-generated FLAIR images approached acquired FLAIR in overall diagnostic quality. In a human-AI collaboration study, BrainGFM assistance yielded a 14% relative improvement in mean radiologist stroke-subtyping accuracy. These findings support generative priors as a reusable foundation for heterogeneous computational brain-imaging applications.