Topological features in addition to radiomics signature predict 1p19q status and tumor grade in low‐grade gliomas

Abstract Radiomics is a procedure to access quantitative features from medical images for predictive analysis. Radiomics is used extensively for malignancy and survival analysis, genomic expression analysis, cancer progression, and assessment. But radiomics does not obtain the connected components, loops, or voids information from a region of interest. Topological data analysis is a newer approach that uses persistent homology to generate multiscale features from an image. We investigated in this study whether the topological features can be as useful as the radiomics features. Glioma grading and 1p19q codeletion assessment requires surgical biopsy, which can be error prone. To reduce biopsy error, MRI‐based radiomics and topological features–based analysis were performed in our study. Topological features produced the best results for 1p19q status and tumor‐grade prediction, with 94.3% (0.93 AUC) and 97.3% (0.97 AUC), respectively, which was an improvement over radiomics. Codes and partial data are available in https://github.com/hellorp1990/TDA-analysis .