Abstract

Diffusion Magnetic Resonance Imaging maps the movement of water molecules, revealing the structure of White Matter (WM). Tractography reconstructs the main WM pathways as 3D curves, referred to as brain fibers. Using cortical parcellation into connected regions is crucial for studying the structural connectome defined by WM connections. Tractography-based parcellation methods aim to define brain parcels by analyzing WM fiber connectivity profiles. However, these methods lack ground truth, challenging their comparison, validation, and improvements. This work proposes a method to generate simulated data as ground truth for evaluating tractography-based parcellation methods. The simulated data consists of random parcels with their connections. First, the method subdivides the cortical surface into random parcels using geodesic distance. Then, it simulates the connections between parcels utilizing a fiber bundle simulator extension based on spline curves, with the ends of the bundles adapted to the shapes of the connected parcels. We used the simulator to generate a simulated dataset of 20 subjects based on a random parcellation, with 150 parcels per hemisphere, which we used to evaluate a tractography-based parcellation algorithm under different input parameters. Results showed up to 118 similar parcels per hemisphere, using a Dice coefficient of 0.5 or higher. This simulator represents the first tool for validating tractography-based parcellation algorithms, offering a realistic and flexible ground truth for their evaluation.