AI and Deep Learning in the Bio-Medical Sciences

Pierre Baldi

We will provide a brief historical overview of AI and deep learning and their two-way interactions with biology. In one direction, progress in deep learning has been driven by progress in neuroscience, from the early beginnings of deep learning to the most recent theories of local learning and the role of gating in attention and transformers. In the other direction, there have been many applications of deep learning to the bio-medical sciences, some of which have also driven progress in deep learning, including the development of convolutional neural networks for fingerprint recognition and of graph/recursive neural networks in protein structure prediction and other molecular applications. We will review such applications from our research group across multiple spatial and temporal scales, including recent applications in biomedical imaging across multiple imaging modalities such as: (1) the analysis of microscopy images in drug discovery; (2) the measurement of vessel calcification or the detection of spinal metastases in radiology;  or (3) the identification of polyps in colonoscopy videos; and discuss future directions.