A temporal argumentation logic for medical diagnosis

Abstract

Clinical decision support systems are systems purposed to improve clinical decision making by matching characteristics of patients to a knowledgebase and generating patient-specific recommendations. The focus of this thesis is on clinical decision support systems that aid clinicians in making diagnoses. More specific, the requirements for such a medical diagnostic system are determined and a logic is proposed on which a medical diagnostic system that meets these requirements can be based.

To determine the requirements for a medical diagnostic system, relevant literature was studied and experts in the medical domain were consulted. It was determined that a medical diagnostic system should be able to handle incomplete information, contradictions and temporal information. Furthermore, it was determined that it is important that such a system allows for the expression of criteria on basis of which certain diagnoses can be excluded. It was established as well that it is important that medical diagnostic programs aid in determining whether a patient has a single disease or multiple ones and that the grounds on which a diagnosis is made can be made clear to users.

According to domain experts, patients are notoriously imprecise in reporting temporal aspects of their symptoms. In addition, temporal information is frequently completely or partially missing for various other reasons. On the other hand, temporal information is in many cases vital for making a diagnosis. Since no logic could be found that meets the requirements stated above and in which partially or completely missing temporal information could be reckoned with, in this thesis a temporal argumentation logic is defined in which it is possible to reason with imprecise and partially missing information.

It was established that a medical diagnostic program based on argumentation logic would meet the most requirements compared to various other logics. The logic proposed in this thesis is based on the argumentation logic DeLP by Alejandro García and Guillermo Simari. The proposed logic distinguishes itself from other prior existing argumentation logics by incorporating a way to express partial temporal information and by providing the possibility to reason with this partial information. In the proposed logic, if the body of a rule contains a temporal constraint and according to the temporal information that is available, it may be possible that this constraints is met, then the head of the rule is derived on basis of the assumption that the temporal constraint are met. Assumptions are tracked very carefully in the proposed logic and it is thus possible to check exactly on which assumptions sentences are derived. In addition, an exact measurement of the strength of assumptions on which sentences are derived is defined.

A medical diagnostic system implementing the proposed logic meets all requirements above to a greater or lesser extent. Furthermore, its feasibility was demonstrated by a proof of concept implementation.