Abstract
This report describes the improvements of a naïve Bayes model that infers the diagnosis of pulmonary nodules in chest CT images based on the findings obtained when a radiologist interprets the CT images. We have previously introduced an inference model using a naïve Bayes classifier and have reported its clinical value based on evaluation using clinical data. In the present report, we introduce the following improvements to the original inference model: (1) the selection of findings based on correlations and the generation of a model using only these findings, and (2) the introduction of classifiers that integrate several simple classifiers each of which is specialized for specific diagnosis. These improvements were found to increase the inference accuracy by 10.4% (p < .01) as compared to the original model in 100 cases (222 nodules) based on leave-one-out evaluation.
