Original Jpn J Radiol Technol 2000 ;
Comparative Performance Study of Several Computer-aided Diagnosis Schemes for Automated Classification of Cardiomyopathy
from Echocardiographic Images

DU-YIH TSAI, ISAMU TAKAI, KATSUYUKI KOJIMA,1) ISAO YAMADA,2) and MASAAKI TOMITA3)
Department of Electrical Engineering, Gifu National College of Technology
1)Department of Administration and Informatics, University of Hamamatsu
2)Department of Radiological Technology, Gifu College of Medical Technology
3)Hirada Hospital
Received Nov. 1, 1999; Revision accepted Jan. 11, 2000; Code No. 590
Summary
The aim of this study was to compare the performance of four different computer-aided diagnosis systems, i.e., a neural network(NN)with back-propagation(BP)learning, an NN with genetic-algorithm(GA)-based learning, a fuzzy logic method, and a GA-based fuzzy logic approach, for automated discrimination of myocardial heart disease. We evaluated the performance of the four systems in terms of accuracy, sensitivity, and specificity. In our experiments, a total of 90 echocardiographic images from 45 subjects(an end-diastole image and an end-systole image from each subject)were used. Four statistical features, namely, angular second moment, contrast, correlation, and entropy were extracted from each composite image obtained from the corresponding end-diastole and end-systole images. These features were subsequently used in our classification schemes. Our results showed that the GA-based fuzzy logic method was superior to the other three methods. This method enabled the classification to achieve 95.9% of the average recognition rate. Thus the GA-based fuzzy logic approach is considered to be a promising tool for the discrimination of myocardial heart disease.
Key words: Computer-aided diagnosis, Fuzzy logic, Genetic algorithm, Neural network, Ultrasonic image