Automated video analysis of immune cell activation and microvascular blood flow in the inflamed microcirculation

The characteristic of blood flow has been shown to be a good indicator of inflammatory conditions. PanagPharma Inc is developing anti-inflammatory agents by studying the effects of synthetic cannabinoids on theblood flow of test animals.. A major bottleneck for this development is the quantitative analysis of blood flowvideos from these experiments. We propose here to develop an automated system based on modern machinelearning methods.We will test two approaches. One will be based on motion segmentation and leveraging previously developedfilters from my lab for microscopic blood analysis. These data are then analyzed with deep convolutionalnetworks. For the second method we will attempt to train a deep network directly on the temporal data. Such anetwork is expected to be more complex than the former. Training such networks is hence more difficult, andwe will therefore use pre-training techniques including unsupervised techniques as well as with simulated data.To generate the simulated data we will develop a unique data simulator that can produce various flowcharacteristics.Deep networks have so far mainly been applied to object recognition in still images, and developing deepnetworks for video processing is a promising research direction. An automated system for the video analysiscan speed up the experimental process and analysis of Panag Pharma's experimental system considerable. Thiswould give Panag Pharma Inc. a cutting edge advantage in this development space and could open a venue fornew developments.