Data-driven decision-making with weights and reliabilities for diagnosis of thyroid cancer

Data science has revolutionized the paradigms of medical decision-making. In the past, medical data could not be recorded and stored indefinitely. In the present day, huge volumes of medical data have been collected electronically, such as medical records, medical images, and heterogeneous surgical data. Under this condition, how to help the radiologists diagnose the thyroid cancer by using the accumulated examination reports and pathologic findings has been a challenge needing to face. From the analysis of historical examination reports, the problem of diagnosing thyroid cancer is evidently considered as a multi-criteria decision-making problem. Thus, a data-driven fusion method of weights and reliabilities in decision-making is proposed in this paper to cope with the above challenge. Linguistic term sets are introduced to model and portray the assessments on each criterion in the problem of diagnosing thyroid cancer by using three types of linguistic scale functions. A data-driven way is then designed to determine the weights and reliabilities of the assessments on each criterion for each radiologist by considering the similarity between the assessments on each criterion and the overall assessments and the similarity between the assessments on criterion and the golden standard, which are derived from the historical data. Subsequently, assessments on each criterion will be combined with the weights and reliabilities to generate a data-driven solution to the problem. The applicability and effectiveness of the data-driven fusion method are verified by solving a real problem of diagnosing thyroid cancer using historical data collected from five radiologists in a tertiary hospital from January 2011 to February 2019.