Design of an optimal planning framework for cryosurgical treatment of brain tumor using CNN segmentation of MRI...

This study presents an integrated methodology for pre-operative cryosurgical planning of irregularly shaped brain tumors using two-dimensional MRI data. Tumor regions were first segmented from MRI slices using a 2D U-Net architecture trained on annotated datasets, enabling accurate geometric reconstruction of the affected tissue. The segmented masks were then employed to determine cryoprobe placement using spatial probe-placement strategies, specifically K-Medoids clustering, Gaussian Mixture Models (GMM), and the Bubble Packing Algorithm (BPA), each of which predicted both the optimal number of cryoprobes and their spatial positioning within the tumor geometry. Subsequently, cryoablation dynamics were simulated in COMSOL Multiphysics using the Finite Element Method (FEM). The thermal model was developed using the Pennes Bio-heat Transfer Equation (PBHTE), with explicit consideration of phase change phenomena through latent heat release to capture the transition from liquid to solid. The objective was to ensure complete tumor coverage by maintaining cytotoxic cryogenic temperatures -40°C to -50°C while minimizing damage to surrounding healthy brain tissue. Results demonstrate improved precision in cryoprobe placement, estimated tumor coverage, and approximate estimation of required freezing times. The proposed framework highlights the potential of integrating deep learning-based segmentation, clustering optimization, and biothermal simulation for effective cryosurgical planning.