A Method for Estimating the Number of Diseases in J-MID Database: Application to CT Report

Abstract

Databases containing medical images play a crucial role in advancing healthcare and broadening our understanding of medical information. They serve essential functions, aiding in diagnosis, facilitating medical training, ensuring quality management, contributing to research on the prevalence of diseases and treatment outcomes. A comprehensive grasp of disease nomenclature within these databases offers a significant advantage, enabling deeper insights into various disease types and distribution patterns. Nevertheless, precisely quantifying the myriad of distinct disease names within image databases poses a significant challenge in numerous healthcare institutions. To address this, our study aimed to estimate disease counts by systematically extracting disease names matching registered medical conditions. This was achieved using a created predicates lexicon specific to disease names in image diagnostic reports written in Japanese. Utilizing this predicate lexicon, our study successfully extracted disease-affirming sentences with high accuracy, demonstrating a sensitivity of 87.9 ± 0.8%, specificity of 93.0 ± 0.6%, and accuracy of 90.5 ± 0.4%. These findings highlight the importance of linguistic tools and systematic approaches in managing medical image databases, ultimately enhancing the efficiency of healthcare processes and the quality of research in the field.