The digital doctor A desirable goal in the field of digital health is to make software understand what doctors know.
- The general practitioner maps arbitrary symptoms to diseases.
- A radiologist can describe a patient’s anatomy based on image modalities, identify suspicious patterns and compare to other patients.
- An oncologist integrates i.a. clinical & biological data, guidelines and the treatment history to make reasonable decisions.
Modelling medical knowledge does not mean to replace physicians. Instead of complete automation, a computer’s inherent advantages over humans (including processing speed, endurance, working memory) are leveraged to build a decision support companion. The physician’s view is hereby extended with relevant information that increases his or her situation awareness and thus enables more effective, efficient and patient satisfactory care.
From an artificial intelligence viewpoint, it is fascinating to identify & train machines comprehensively in such a complex, dynamic & information-rich domain. Technically, it is challenging to integrate the vast amount of pattern recognition & combinatory knowledge that doctors possess. It must be divided and organized in interdependent data mining tasks without the need of laboriously labelled massive datasets. Physicians rather state their interest in a particular information by providing few examples from which the software learns. One-Shot Learning, automated machine learning as well as techniques inspired by cognitive science will play important roles in this vision.