Algorithm and Structure of Parallelization of 3D Spline-Based Segmentation Processes of Glioblastoma MRI Images

Glioblastoma multiforme (GBM) is one of the most aggressive and common malignant brain tumors of the central nervous system. It is characterized by rapid infiltrative growth, heterogeneous structure, and unclear boundaries, which make accurate segmentation difficult. MRI (Magnetic Resonance Imaging) images are the main diagnostic tool in the diagnosis of glioblastoma, and the data obtained in 3D form allow for an accurate assessment of the size, shape, and structure of tumors. In this paper, glioblastoma segmentation based on 3D spline interpolation is performed, and the process is completed by feature extraction based on PyRadiomics and classification into types using the RandomForest classifier. As a major innovation, a parallelization algorithm accelerates the calculation of 3D spline points and the segmentation process.
During the research, pre-processing (normalization, noise reduction) of MRI volumetric data, automatic ROI (Region of Interest) segmentation using Otsu's method, and 3D contour drawing based on spline were performed. The parallel computing engine was created using Python's concurrent.futures module, and the feature extraction for each patient's MRI volume was performed in separate processes. The results showed that the computation time was approximately 4.3 times faster on 8 processor cores.
The proposed approach not only improved the segmentation accuracy (F1-score = 0.929), but also significantly improved the computational efficiency. This algorithm is suitable for application in medical diagnostic systems, especially in clinical applications requiring real-time operation.