Kavli Affiliate: Saul Kato
| Authors: Jason Adhinarta, Jizheng Dong, Tianxiao He, Daniel Sprague, JIa Wan, Hyun Jee Lee, Zikai Yu, Hang Lu, Eviatar Yemini, Saul Kato, Erdem Varol and Donglai Wei
| Summary:
The nematode C. elegans is a well-studied model organism for characterizing the structure, connectivity, and function of a complete nervous system. Recent technical breakthroughs in 3D light microscopy and fluorescent protein tagging of individual neurons have brought us closer to capturing the neural dynamics of the worm at whole-brain resolution. Nevertheless, capturing a complete map of neural dynamics using these high-resolution recordings requires solving three specific challenges: i) detection of individual neurons in fluorescence videos, ii) identification of these neurons according to their anatomically defined classes, and iii) tracking of neural positions over time. Successfully addressing these challenges with high sensitivity, specificity, and throughput can enable us to analyze a large population sample, providing unprecedented insights into the structure and function of an entire brain at single-neuron resolution—a feat previously unaccomplished in any organism. To facilitate this scientific goal, we have curated publicly available annotated datasets from 118 worms across five distinct laboratories and established systematic benchmarks, decomposing the overarching objective into three well-defined tasks: i) neural detection, ii) identification, and iii) spatiotemporal tracking. Our preliminary analysis has revealed considerable room for improvement in existing state-of-the-art computational methods. Consequently, we envision that our WormID-Benchmarkcan catalyze efforts by a broad audience specializing in computer vision to develop robust and accurate methods that significantly enhance the throughput of generating annotated whole-brain neural dynamics datasets. We make our benchmark results reproducible; our code is publicly available at https://github.com/focolab/WormND.