A Method for Estimating the Number of Diseases in an Image Database: Utilization of Predicates and Application to a CT Database.

Medical image databases are crucial in the advancement of healthcare and acquisition of knowledge. They support diagnosis, training, quality management, and research on disease prevalence and treatment outcomes. Understanding disease names within these databases provides the advantage of comprehensively grasping the types and distribution of diseases. However, accurately determining the number of diseases within image databases poses a challenge in many hospitals. Therefore, this study aimed to estimate the number of diseases by extracting disease names matching registered diseases and utilizing a lexicon of positive or negative predicates for disease names in Japanese image diagnostic reports. With the created predicate lexicon, we were able to extract sentences affirming the presence of diseases with high accuracy (sensitivity = 81.0 ± 5.4%, specificity = 86.2 ± 5.0%).