In the modern healthcare system, medical imaging technologies, such as CT, MRI, Ultrasound, histology images, fundus photography, play important roles in disease diagnosis, assessment, and therapy. Deep learning, a subfield of AI, has seen a dramatic resurgence in the recent few years, largely driven by increases in computational power and the availability of massive new datasets. Because of the increasing proliferation of medical devices and digital record systems, computer-aided diagnosis stands to benefit immensely from deep learning, which could aid physicians by offering second opinions and flagging concerning areas in images. In this talk, I will present a series of deep learning methods towards interdisciplinary researches at AI for medical image diagnoses, such as cancer classification and segmentation, anatomy tissue semantic parsing, and rare disease prediction. The proposed techniques cover a wide range of deep learning topics including neural network architecture design, semi-supervised and unsupervised learning, multi-task learning, few-shot learning, etc. The challenges, up-to-date progress, and promising future directions of AI-powered healthcare will also be discussed.