Duke University's AI algorithm maps neuron activity 50x faster than humans

Researchers at Duke University have developed an artificial intelligence algorithm that dramatically speeds up the analysis of neuron activity in mouse brains. Using two-photon calcium imaging, scientists track individual neurons as they fire, but manual segmentation of these videos is painstakingly slow—taking 4 to 24 hours for a single 30-minute recording. The new deep learning-based system, described in the Proceedings of the National Academy of Sciences, automates this process with equal or better accuracy, completing the same task in just 20 to 30 minutes.

Lead author Somayyeh Soltanian-Zadeh stated the algorithm is 'as accurate as, if not better than, human experts.' The system was trained on diverse datasets to generalize across different brain layers and neuron sizes. The tool is now publicly available, allowing neuroscience labs worldwide to accelerate brain mapping studies. This advancement could unlock faster insights into neural circuits, learning, and neurological disorders, removing a key roadblock in understanding how the brain works.