Detection of glioma evolution on longitudinal MRI studies

Detection of millimetric brain tumor growth patterns on longitudinal MRI acquisitions remains challenging in clinical practice. A simple difference map of two longitudinal co-registered MRI volumes fails to detect specific tumor evolution, due to non-linear contrast change between the two data sets. This paper presents a novel method for detection and quantification of tumor evolution in longitudinal single-protocol MRI studies. A computational framework was designed to enable comparison of co-registered MRI volumes based on gray-scale "normalization" via midway histogram equalization and computation of difference maps. Midway-based difference maps provided very selective representations of structural modifications within pathological areas and on the surrounding structures. Quantitative tumor growth parameters between times t(1) and t(2) were computed on the difference maps, provided that a manual segmentation of the tumor is available at time t1. The method was evaluated on longitudinal SPGR (T1-weighted) and FLAIR (T2-weighted) MRI volumes for two patients harboring a WHO grade II glioma. Results for quantification of tumor growth from midway difference maps are presented, showing sub-millimetric precision of clinical growth indices, when compared to manual tracing estimations.