Case Study: MIT accelerates serotonin neuroscience research and saves thousands of hours with Labelbox

How MIT researchers are using neural nets to automate manual tasks for novel serotonin research

MIT researchers led by PhD student Jungsoo Kim explored how serotonin modulates neural circuits and collected terabytes of microscope and behavior images of C. elegans. They needed high‑quality labeled data to train and benchmark neural nets, but manual annotation and traditional computer‑vision tools were too slow and pre‑trained models were not suited to their specialized images. To distribute annotation across the group and manage data and APIs, they selected Labelbox (using the Labelbox Workforce service) as their web‑based labeling platform.

Using Labelbox, the team created an object‑detection workflow to mark worm centroids, labeled ~2,000–3,000 images, and enforced multipass review (one third labeled three times) with automated quality checks. Labelbox’s platform and Workforce produced fast turnaround (often within 24 hours) and high-quality annotations—98% of multipass labels met predefined criteria—enabling neural nets to replace manual pipelines for tracking, segmentation and detection, saving the team thousands of hours and accelerating their research.

MIT