3D MRI Brain Tumor Diagnosis with Topological Descriptors

In this paper, we introduce a novel approach that utilizes topological data analysis (TDA) in conjunction with MRI for determining glioma grade and identifying genomic biomarkers. The study hypothesizes that by examining the evolution of topological patterns across various grayscale values, it is possible to identify distinct topological footprints left by different tumor classes in MR images. These footprints can be used as powerful feature vectors for tumor classification. The results of the study demonstrate that higher dimensional topological features provide a powerful ML model, achieving toe-to-toe results with the existing state-of-the-art models in accurately classifying both low- and high-grade gliomas. Additionally, the proposed method achieves high accuracy in predicting the methylation status of the MGMT promoter, which is crucial for prognostic assessment and treatment response prediction. By incorporating TDA output, this research has the potential to advance MRI-based deep learning approaches, enabling the identification of clinically significant tumor features and facilitating informed clinical decision-making.