Physics-informed UNETR: A Brain Tumor Segmentation Combining UNETR Architecture and Physics Constraints Using Brats2021 Dataset and Insights for Molecular Imaging

Abstract

This Physics-Informed UNETR model incorporates both learned features from MRI scans and the physical constraints of tumor dynamics, offering a more holistic and accurate segmentation. By integrating reaction-diffusion equations, which govern tumor expansion, the model aims to provide more accurate segmentation results, especially in complex cases with irregular tumor boundaries. The problem, therefore, lies in how to effectively combine state-of-the-art deep learning techniques with well-established biological models to improve the accuracy and reliability of brain tumor segmentation. This study seeks to address this gap by developing a novel approach that leverages both data and physics, providing a more informed and robust solution to the brain tumor segmentation problem. The physics models incorporated could allow us gain molecular insights into brain tumors especially cell metastatic processes via diffusion and cell transport.