Grant: Project Grant

A leading theme in human stroke treatment are means of rehabilitation that engage surviving brain circuits to compensate for damage due to stroke. However, these approaches still require optimisation in animal models. We suggest that the study of stroke will benefit from advances in computer science methods: including computer vision where software can see features too subtle or fast for human raters, and machine learning where computer programs can be advanced without explicit instructions. Here we employ the use of a synthetic animated mouse to create realistic videos that are used for training advanced computer models (through a trick borrowed from "deepfakes"). The advantage of the model is that it can simulate aspects of mouse behaviour that would otherwise be difficult to observe in nature (due to a rare occurrence, poor timing, or inability to resolve in 3 dimensions (3D)). These trained computer models are then used with real video data to more carefully evaluate stroke disease progression and recovery by representing mouse movements within 3D. The animated mouse interface also provides a common system to represent mouse movements enabling pooling of data leading to statistically stronger experiments. Computer vision combined with the online analysis of brain activity offers the promise of rapidly cataloging specific mouse movements in such a way that stroke recovering animals can be subjected to corrective rehabilitation. By having the ability to monitor complex movements such as reaching, as well as the brain activity associated with planning and executing them we can give the animals real-time feedback about their progress in recovery. Forms of rehabilitation would be instituted in a closed-loop manner (where results are fed back) and reward animals which are moving towards recovery in both their affected brain circuits and specific movements. We aim to determine rules for how these approaches can be optimally applied to promote recovery from stroke.