Accelerating Neuron Reconstruction with PATHFINDER

Abstract

Comprehensive mapping of neural connections is essential for understanding brain function. Existing automated methods for connectome reconstruction from high-resolution images of brain tissue introduce errors that require extensive and time-consuming manual correction, a critical bottleneck in the field. To address this, we developed PATHFINDER, an AI system that segments volumetric image data, identifies potential ways to assemble neuron fragments, and evaluates the plausibility of resulting shapes to reconstruct complete neurons. Using a dataset of all axons in an IBEAM-mSEM volume of mouse cortex, we show that PATHFINDER reduces the error rate in axon reconstruction by an order of magnitude over previous state of the art, leading to an improvement in proofreading throughput of up to 84x relative to prior estimates in the context of a whole mouse brain. By drastically reducing the manual e"ort required for analysis, this advance unlocks the potential for both large-scale connectome mapping and routine investigation of smaller volumes.