Predicting of Aneurysm with Learning Vector Quantization in Patients with Kawasaki Disease

ABSTRACT

PURPOSE: We applied Learning Vector Quantization (LVQ) in the analysis of data from Kawasaki disease patients with coronary artery aneurysm in an attempt to achieve accurate predictions of outcome for individual patients. METHODS: One hundred and seventy-five patients with Kawasaki disease were recruited. First, data of 75 patients (of which 60 patients had no aneurysm and 15 patients had aneurysm) were trained using the network. The data contained age, sex, white blood cell count, platelet count, CRP, and serum albumin of each individual. Then, data of another 100 patients (of which 83 patients had no aneurysm and 17 patients had aneurysm) were integrated into the previously trained LVQ neural network. After the analysis was completed, the sensitivity, specificity, positive predictive value, and negative predictive value were compared between the LVQ neural network and Harada risk index. RESULTS: Out of 100 patients, it was evaluated that 21 patients had coronary aneurysm and 79 patients did not. Among the 21 patients, 15 patients actually did have the disease, while 77 out of 79 patient actually did not have coronary aneurysm. The sensitivity, specificity, positive predictive value and negative predictive value were 88.2, 92.7, 71.4 and 97.4%, respectively. The values by the conventional Harada scoring were 52.9, 92.7, 60.7 and 91.7%, respectively, and contrasting the values above. CONCLUSOIN LVQ neural network can classify and predict patients who, in the future, will have coronary artery aneurysm. This method is superior than the conventional methods in predicting aneurysm.