DISCo: Deep learning, Instance Segmentation, and Correlations for cell segmentation in calcium imaging

• DOI: 10.48550/arxiv.1908.07957
• Publication Date: 2019-01-01
• Authors: Elke Kirschbaum; Alberto Bailoni; Fred A. Hamprecht
• Keywords: Calcium imaging; Benchmark (surveying); Neurophysiology; Grid cell

Calcium imaging is one of the most important tools in neurophysiology as it enables observation neuronal activity for hundreds cells parallel and at single-cell resolution. In order to use data gained with calcium imaging, necessary extract individual their from recordings. We present DISCo, a novel approach cell segmentation videos. temporal information recordings computationally efficient way by computing correlations between pixels combine shape-based identify active well non-active cells. first learn predict whether two belong same cell; this summarized an undirected, edge-weighted grid graph which we then partition. so doing, approximately solve NP-hard correlation clustering problem recently proposed greedy algorithm. Evaluating our method on Neurofinder public benchmark shows that DISCo outperforms all existing models trained these datasets.