Addressing Pathological Variability in Skull Stripping Volumetric Brain MRI

Abstract

"Magnetic Resonance Imaging (MRI) is a critical tool for brain imaging, offering detailed views of cerebral anatomy necessary for diagnosing a variety of neurological conditions. Skull stripping, the process of isolating brain tissue from other structures in Brain MRI scans, is essential for accurate analysis. However, current automated methods often struggle with MRIs that exhibit pathological changes near the skull, a common occurrence in clinical practice. This study introduces a 3D Complementary Segmentation Network (CompNet) designed to enhance the skull stripping process, particularly for Brain MRIs with such complex pathologies near the skull boundary. The proposed architecture integrates dual complementary pathways that concurrently learn to identify brain and non-brain tissues, improving the distinction even in the presence of brain pathologies that challenge traditional and existing automated methods. The enhanced model shows notable performance on 3D T1-weighted Brain MRIs, particularly on scans affected by tumors and other abnormalities near the skull boundary. suggesting its potential for widespread clinical adoption in neurological diagnosis and treatment planning. Subject Descriptors: Skull Stripping, Brain Magnetic Resonance Imaging, Image Segmentation, MRI Analysis, Neural Network Architectures Machine Learning in Medicine Biomedical Engineering"