MegaTrack: a framework for the anatomically accurate and time-efficient virtual dissection and analysis of large-scale tractography datasets

Abstract

The increasing prevalence of large-scale neuroimaging initiatives has created a need for efficient and anatomically accurate methods for analysing tractography data. Here, we present MegaTrack, a framework designed to combine the anatomical precision of manual dissections with the efficiency required for big data analysis. At its core, MegaTrack operates by normalising individual tractography datasets to a standard space, concatenating them into a single "mega" tractography dataset the user manually dissects before automatically extracting subject-specific tracts and metrics in native space. This approach permits identifying and extracting tracts not yet available in existing templates or automatic dissection tools. We validated this approach using multiple datasets and applications. Comparison with individual manual dissections showed high spatial agreement (weighted dice coefficient > 0.95) and strong correlations in tract-specific metrics (R2 > 0.9). In a longitudinal dataset, MegaTrack demonstrated improved reproducibility across time points and higher inter-rater reliability compared to manual dissections. In a motor neuron disease cohort, MegaTrack replicated known group differences in corticospinal tract microstructure with comparable sensitivity to manual dissections (p = 0.001) but with significantly reduced processing time. Finally, we showcase MegaTracks scalability by creating a comprehensive white matter atlas from 140 healthy adults and demonstrate its application for lesion analysis in stroke patients. This framework is accompanied by an interactive online tool that allows users to explore these atlases and perform customised lesion analyses. The MegaTrack framework is a time-efficient solution for the anatomically precise analysis of large tractography datasets, enabling the rapid simultaneous dissection of multiple subjects while preserving individual anatomical features. The frameworks ability to efficiently process large datasets makes it an excellent tool for generating training data for machine learning and deep learning algorithms in tractography analysis. This approach has broad applications in both research and clinical settings, from group comparisons to personalised lesion mapping, and can significantly contribute to the development of advanced automated tractography methods.