Abstract 15136: Toward the Development of an AI Application to Diagnose Ischemic Heart Disease From a Medical Interview

Abstract

Background: Interviewing technique for patients with chest pain depends on each doctor’s skill. Therefore, an artificial intelligence (AI)-based interview system may be useful to stratify probability of ischemic heart disease based on patient’s complaint. Methods: As a definitive diagnosis corresponding to the interview is needed as a sample, a medical interview is taken from patients presenting with chest pain as a chief complaint, which is combined with the patient's baseline data to develop an AI algorithm. Results: The items used as input attributes in the interview are listed in Table 1. A multilayer perceptron (MLP) was applied to the subjects (N=125) for the prediction of patients diagnosed with angina pectoris (54 cases). These models were validated using the leave-one-out method. Comparisons were made with models in which each of diabetes (45 cases), dyslipidemia (88 cases), hypertension (95 cases) and smoking (never: 60 cases; past: 50 cases; current: 15 cases) were added as attributes in addition to the items in Table 1. (Table 2) The ROC curves of the multilayer perceptron with only the questionnaire items in Table 1 as input attributes and the multilayer perceptron with the dyslipidemia attribute added are shown in Figure 1. Conclusion: It was found that the diagnosis of angina pectoris could be detected interview data while the prediction accuracy was improved by adding the attribute of dyslipidemia. Further investigation is needed to complete a highly accurate AI-based application by increasing the number of samples.