A Case Study of a Machine-Learning Approach in Differential Diagnosis of Schizophrenia: The Predictive Capacity of WAIS-IV / 신경정신의학

OBJECTIVES: Machine learning (ML) encompasses a body of statistical approaches that can detect complex interaction patterns from multi-dimensional data. ML is gradually being adopted in medical science, for example, in treatment response prediction and diagnostic classification. Cognitive impairment is a prominent feature of schizophrenia, but is not routinely used in differential diagnosis. In this study, we investigated the predictive capacity of the Wechsler Adult Intelligence Scale IV (WAIS-IV) in differentiating schizophrenia from non-psychotic illnesses using the ML methodology. The purpose of this study was to illustrate the possibility of using ML as an aid in differential diagnosis. METHODS: The WAIS-IV test data for 434 psychiatric patients were curated from archived medical records. Using the final diagnoses based on DSM-IV as the target and the WAIS-IV scores as predictor variables, predictive diagnostic models were built using 1) linear 2) non-linear/non-parametric ML algorithms. The accuracy obtained was compared to that of the baseline model built without the WAIS-IV information. RESULTS: The performances of the various ML models were compared. The accuracy of the baseline model was 71.5%, but the best non-linear model showed an accuracy of 84.6%, which was significantly higher than that of non-informative random guessing (p=0.002). Overall, the models using the non-linear algorithms showed better accuracy than the linear ones. CONCLUSION: The high performance of the developed models demonstrated the predictive capacity of the WAIS-IV and justified the application of ML in psychiatric diagnosis. However, the practical application of ML models may need refinement and larger-scale data collection.

Implementation of Medical Diagnostic System Based on Epidemiological Data / 대한의료정보학회지

OBJECTIVE: We implemented automatic online medical consultation software. It infers disease of patients with knowledge about symptoms and the epidemiologic data. And we compared its performance of inference with that of human doctors. METHODS: This software accepts information about users' age, sex, and symptoms, lists up diseases compatible with these information, and sorts diseases by probability. We implement this software with Ruby and C. RESULTS: We compared diseases listed up by this software with those that by two human doctors, and found that 1) 90% of confirmed diagnoses was included in the list this software inferred, and 2) more than 50% of diseases in the list this software inferred are same diseases as ones both of two human doctors inferred. CONCLUSION: This software can not determine final diagnosis. But this software lists up probable diseases only by interview. Then we believe this software will be useful for patients when they want to check themselves before consulting their doctor. We believe that this software will be useful for patients to check their health status.