BossDB Kasthuri Challenge

Background

Large-scale, high-resolution neuroanatomical imaging produces vast volumes of data that offer new opportunities to understand brain structure and connectivity. Machine and deep learning methods are increasingly used to analyze these datasets, but standardized tasks, annotations, and benchmarks are still needed to compare methods and drive progress.

The Kasthuri Challenge provides a standardized segmentation and detection benchmark on a saturated reconstruction of a sub-volume of mouse neocortex imaged with a scanning electron microscope (SEM). The primary targets are synapse and membrane annotations—regions that best reveal microstructural connectivity and provide robust targets for segmentation algorithms. This project adapted ideas from prior multitask neuroimaging benchmarks to create a reproducible challenge and baseline code for the community.