Deep Interactive ۳D Visualization of Brain Tumors Using MRI Reconstruction and Augmented Reality via Spatial-Attention U-Net Optimized with Multi-Objective Genetic Algorithm

Precise visualization of brain tumors is critical for enhancing neurosurgical planning and intra-operative decision-making. This study presents a novel deep-learning-based framework for accurate brain tumor segmentation, ۳D reconstruction, and augmented reality (AR)-based visualization. The model utilizes the BraTS۲۰۲۱ dataset comprising multimodal MRI sequences (T۱, T۱CE, T۲, FLAIR). The preprocessing pipeline integrates Bilateral filtering and Histogram Equalization techniques to suppress noise and improve contrast. For segmentation, a customized U-Net architecture incorporating a Spatial Attention module is implemented to emphasize relevant image regions. Additionally, a multi-objective Genetic Algorithm is employed to optimize network parameters, achieving a balance between segmentation accuracy (Dice coefficient), computational cost, and training time. The resulting segmentation masks are reconstructed into volumetric ۳D models using MATLAB graphical functions such as isosurface and vol۳d, exported in standard ۳D formats (OBJ/STL), and rendered in an Android-based AR environment using Google ARCore. This system supports interactive, real-time visualization of brain tumor morphology, providing a valuable tool for preoperative assessment and neurosurgical education. Results suggest that the proposed pipeline can improve the interpretability and precision of tumor localization compared with conventional MRI visualization. Future work includes cross-validation using the BraTS۲۰۲۵ dataset and integration with intraoperative navigation systems.

کلیدواژه ها:

Brain tumor segmentation ، ۳D reconstruction ، Augmented realityDeep learning ، U-Net ، Genetic algorithm ، Spatial attention ، MRI.