EFFECTIVE SKULL STRIPPING AND MULTI-TISSUE BRAIN SEGMENTATION FOR EPILEPSY DIAGNOSIS

The MRI-based diagnosis of epilepsy relies on the detection of subtle changes in the brain, which
requires exceptionally clean and refined image preprocessing. One of the major challenges is the robust
removal of non-brain tissue and the proper delineation of brain tissue. To address variability in skull
stripping, FreeSurfer’s SynthStrip is used, executed within Docker Desktop to ensure cross- platform
consistency and reproducibility. For the segmentation of white and gray matter from T1-weighted MRI
images, FastSurfer is a deep learning-based neuroimaging pipeline used for the fully automated
processing of structural human brain MRIs. By leveraging FastSurfer's advanced Convolutional
Neural Network architecture rather than traditional clustering, the proposed approach maintains high
spatial resolution and structural fidelity even when images exhibit varying levels of contrast. Validating
this framework on a wide variety of T1-weighted image data, the skull-stripping achieved a Volume
Reduction Ratio of 80.12%, where the average VRR is 47.2%, and the proposed segmentation attained
the dice score of 91%. In conclusion, the proposed segmentation can be efficiently used for the detection
of neurological disorders.