HD-BET

HD-BET uses neural networks to separate brain tissue from surrounding structures in MRI scans. It runs locally and is aimed at researchers and medical imaging teams who need automated brain masks, including for scans affected by tumors, lesions or surgical resection cavities.

Its training covers precontrast and postcontrast T1-weighted images, T2-weighted images and FLAIR. Each sequence can be processed independently, so its use isn't limited to precontrast T1 scans. The training data came from a multicenter clinical trial in adults with brain tumors and included different scanners, acquisition settings and treatment-related tissue changes.

HD-BET was also evaluated on the NFBS, LPBA40 and CC-359 benchmark datasets. In the reported comparisons, it achieved better brain-mask overlap and boundary accuracy than FSL BET, AFNI 3DSkullStrip, Brainsuite BSE, ROBEX and BEaST. Its neural network implementation builds on nnU-Net.

The Python tool is open source under Apache 2.0 and has been extensively tested on Linux; Windows and macOS are expected to work too. It supports GPU acceleration, Apple's MPS backend and CPU processing. CPU runs take longer and need substantial RAM. Model parameters are downloaded to the user's machine.

HD-BET accepts 3D NIfTI files and can process whole folders of scans. It produces brain masks for each input. It doesn't directly accept 4D image sequences, and input scans must match the orientation of the standard MNI152 template.