Publication

Abstract : The limited amount of medical datasets continues to be a major bottleneck in medical image processing, despite the remarkable success deep learning models like CNNs have had in this field. In an effort to tackle this issue, scientists have begun searching outside of the currently accessible medical datasets for additional information. Conventional methods often use transfer learning to extract information from natural photos. More recent efforts use the domain expertise of physicians to build networks that imitate their diagnostic patterns, mirror their training processes, or concentrate on the characteristics or regions that physicians find particularly important. The present state of research on incorporating medical domain knowledge into deep learning models for a range of tasks, including illness diagnosis, lesion, organ, and anomaly detection, as well as lesion and organ segmentation, systematically classify various medical domain knowledge types used for each task along with matching integration techniques. They also offer prospective research directions and difficulties that are now being faced.