Improved statistical classification methods in computerized psychiatric diagnosis

Background. Mainstream psychiatric diagnosis involves mainly sequential, ex pert-system-derived, logical decision rules. Among the few statistical clas sification methods that have been sporadically evaluated are Bayes, k-neare st neighbor, and discriminant analysis classifiers. Methods. A statistical classification method based on artificial neural networks (ANN) with task-s pecific constrained architectures was applied to a sample of 796 clinical i nterviews, where the symptom evaluation and the diagnostic judgments were m ade using the Psychiatric State Examination (PSE) system. The proposed cons trained ANN (CANN) method was compared with other statistical classificatio n methods. Results. CANN was found to be superior to all other considered m ethods, having an overall "correct" classification rate of 80% when applied to test data. Similarly, the concordance coefficients of agreement with th e PSE diagnostic categories were all very high. Among the other used method s, discriminant analysis had slightly inferior performance but better gener alization capability. Conclusions. The proposed CANN method has a definite utility in psychiatric diagnosis and requires further evaluation, perhaps a longside other standard classification systems and/or with larger samples.