The use of modified constellation graph method for computer-aided classification of congenital heart diseases

The authors describe a novel method of data reduction and classification in a multidimensional symptom space for use as a diagnostic aid for congenital heart diseases. The algorithm developed here reduces interactively a multidimensional symptom space to sectorial regions representing each disease in a semicircle using the modified constellation graph method. This method enables classification of patients, using the angle in the semicircle as a single classifying parameter with an accuracy of about 90%, with little overlapping between disease sectors. Comparing this method with conventional factor analysis, the authors found the former far more effective than the latter for disease region separation.