HD-BET

HD-BET performs automated brain extraction from multisequence magnetic resonance imaging (MRI) to provide accurate brain masks for preprocessing of heterogeneous and pathologically altered neuroimaging datasets.

Key Features:

• Artificial Neural Networks: Employs artificial neural networks to perform brain extraction across diverse and complex MRI sequences.
• Rigorous Validation: Validated against six popular, publicly available brain extraction algorithms across several large-scale neuroimaging datasets, including a prospective multicentric trial in neuro-oncology.
• Performance Metrics: Shows median improvements of +1.16 to +2.50 points for the Dice coefficient and reductions of -0.66 to -2.51 mm for the Hausdorff distance.
• Robustness Across Variability: Robust to pathology- or treatment-induced tissue alterations and consistent across a broad range of MRI sequence types, hardware, and acquisition parameters.

Scientific Applications:

• Preprocessing for downstream analyses: Provides accurate brain masks to improve reliability of subsequent neuroimaging analyses.
• Clinical neuro-oncology: Supports processing of pathologically altered brains, expanding applicability in neuro-oncology imaging studies.
• Multicenter and heterogeneous dataset processing: Enables automated extraction across heterogeneous multisequence MRI datasets and varying acquisition settings.

Methodology:

Uses artificial neural networks for brain extraction and was evaluated by comparing against six publicly available brain extraction algorithms using Dice coefficient and Hausdorff distance on multiple large-scale neuroimaging datasets including a prospective multicentric neuro-oncology trial.