Dempster-Shafer Belief Structure And Aggregations - Usage in Medical Diagnosis

Author: Meri Chonishvili
Keywords: multi criteria decision making, Dempster-Shafer belief strcture, fuzzy measure, aggregation methods, q-ROFN, Choquet integral, discrimination analysis, Dempster's extreme expectations, medical diagnosis
Annotation:

When diagnosing a medical condition, usually multiple symptoms and their combinations must be considered. Each symptom and combination can be assigned a weight indicating its importance to the diagnosis. We can note certain similarity between medical diagnostic and the process of decision making using Dempster-Shafer belief structure. Diagnoses are considered as possible alternatives and symptoms – as criteria (or attributes). Then sets of related symptoms can be viewed as focal elements, and the importance of these sets – focal probabilities (basic probability assignments). Once medical diagnosis problem is transformed into the framework of Dempster-Shafer belief structure, we may use multiple aggregation methods to sort alternatives and decide on the most likely diagnose. In addition, using q-rung ortho-pair fuzzy numbers for providing initial data will give experts more freedom in expressing their evaluations and attitudes. As for the ordering of alternatives, it can be greatly affected by the choice of aggregation method. The main objectives of this paper are to incorporate q-rung fuzzy numbers in the process of decision making using Dempster-Shafer belief structure and tailor the process onto medical diagnosis problem. When making so we will discuss multiple aggregation methods – Choquet integral and novel discrimination analysis method among them. To demonstrate results of this paper, a simple web-application has been implemented. It consists of an admin panel and a REST-API. The panel enables user to configure components of Dempster-Shafer belief structure. The API has endpoint for loading that data as a JSON and an endpoint for running the decision-making process. The latter returns sorted alternatives with corresponding aggregated values as well as optimal alternative for each aggregation method. The API is authenticated via bearer token and can be used in any REST-based application. The results of this paper can be further developed in cooperation with medical experts. While problem discussed here is a hypothetical one, real scenarios and data can be acquired from the healthcare providers. This will make it possible to create a system which provides reliable recommendations in diagnosing medical conditions.

Lecture files:

დემპსტერ-შეიფერის მონაცემთა ტანის გამოყენება სამედიცინო დიაგნოსტიკაში - პრეზენტაცია [ka]