Explanation models as a component of the intelligent computer-aided diagnosis systems in medicine: a brief review

Abstract

The paper considers the most important and effective approaches and models for explaining and interpreting diagnostic results obtained using intelligent computer-aided diagnosis systems. The need to use them is due to the fact that the intelligent computer-aided diagnosis system itself is a “black box” and it is important for the doctor not only to get the patient’s diagnosis, but also to understand why such a diagnosis is stated, what elements of the patient information are the most significant from the point of view of the diagnosis. Reviews of the main approaches to explain predictions of machine learning models applied to general areas as well as to medicine are presented. It is shown how different types of the initial patient information impact on the choice of explanation models. Models are considered when visual or tabular information is available. Example-based explanation models are also studied. The purpose of the work is to review the main explanation models and their dependence on types of information about the patient.