QUANTITATIVE ASSESSMENT OF MYOCARDIAL ULTRASOUND TISSUE CHARACTERIZATION THROUGH RECEIVER OPERATING CHARACTERISTIC ANALYSIS OF BAYESIAN CLASSIFIERS

Objectives. This work proposes a self-consistent assessment methodolog y for quantitative evaluation of any combination of diagnostic feature s, with the immediate goal of quantitatively assessing the discriminat ing power in diabetic patients of features derived from ultrasound bac kscatter from myocardium. Background. Four features from analysis of l eft ventricular myocardial ultrasound backscatter have previously been shown to be sensitive to potentially cardiomyopathic changes in patie nts with insulin dependent diabetes mellitus who have no overt heart d isease. The measured features were significantly different between suc h patients and normal control subjects, as well as among groups of suc h patients with and without systemic complications of the disease. The quantitative discriminating potential of the features was not assesse d. Methods. Multivariate classifier functions were constructed and ana lyzed by using the methodology of the receiver operating characteristi c curve, which allows quantitative assessment of the discriminating po wer of these features, alone or in combination. The area under the rec eiver operating characteristic curve-the true positive rate averaged o ver all false positive rates-was used as a summary measure of performa nce. Results. In distinguishing patients with insulin-dependent diabet es mellitus from normal control subjects, the most discriminating comb ination of ultrasound features for the detection of such changes in th ese patients yielded receiver operating characteristic curves with are a measures of similar to 0.80; for such patients with retinopathy the measure increased to 0.90. This performance is comparable to that of m any commonly used diagnostic tests. Conclusions. A self-consistent set of evaluation methodologies has quantitatively demonstrated the sensi tivity of four ultrasound backscatter features to otherwise latent cha nges in myocardial structure that accompany the evolution of insulin-d ependent diabetes mellitus. The results are remarkable in themselves a nd suggest the potential of the features for tbe general field of card iac ultrasound tissue characterization.