Recent advances in Computer Vision (CV), especially in deep learning and Machine Learning (ML) have the potential to transform clinical research and health care. Over the last decade, my group has developed deep learning systems for the detection and recognition of faces, emotions, objects and actions, as well as CV algorithms for markerless motion capture and gait analysis – this has resulted in dramatic performance on unconstrained image and video data. In addition, we have shown the effectiveness of Generative Adversarial Networks (GANs) for domain adaptation with applications in semantic segmentation and object recognition. We have also studied the problem of forensic examiners and deep learning algorithms working together. In this talk, I will describe some of these developments and discuss potential applications of deep learning in the health sciences, including multi-modal data fusion, prediction of undesirable outcomes from medical signals and images, evaluating pain based on facial expressions and monitoring patients (e.g. video EEG) and their movements. While the prospects of ML for transforming health care and clinical research are exciting, several challenges such as learning from smallannotated data or large unlabeled data, adapting/generalizing ML algorithms to related and novel tasks and designing communicable ML techniques remain to be addressed.
Read More