Brain Logistic Segmentation (BLS): an efficient algorithm for whole-brain tissue segmentation in structural magnetic resonance imaging

Purpose Structural magnetic resonance imaging (MRI) data is essential for many neuroscience and clinical applications. The morphological features presented in the human brain are a rich source of information for understanding healthy development and pathological status, assisting the evaluation of brain atrophy, cortical thickness, and healthy brain aging. Several efforts have been made by the scientific community to offer robust automatic segmentation methods to brain analysis, helping health professionals in such complicated and time-consuming tasks. However, the precise definition of brain tissues is not trivial and still is an open problem, being an important topic for modern research. Methods We presented an improved brain segmentation algorithm in this study, which is focused to robustly detect the main brain tissues, i.e., CSF, GM, and WM. We propose the Brain Logistic Segmentation (BLS) method based on multiple logistic classification algorithms, which was evaluated with simulated (BrainWeb) and real MRI image data sets (IBSR, OASIS). The brain segmentation quality was measured with Dice (DICE), Volume Similarity (VOLSIM), the 95th percentile of Hausdorff Distance (HDIST), normalized Whole-Brain Volume Absolute Error (nWBV-AE), and compared speeds with total processing time (t). Results We compared the proposed method (BLS) to the most well-known brain tissue segmentation algorithms, i.e., FSL-FAST, Atropos, SPM12, and BrainSuite-PVC. The results show a significant ( p < 0.01, paired t test) improvement in segmentation with the BLS method, evidencing a better local tissue definition with the BLS. Conclusion The BLS method showed significant improvements to be eligible for brain analysis in neuroscience and clinical applications such as brain atrophy and cortical thickness measurement.