CLASSIFICATION OF LOWER-LIMB ARTERIAL STENOSES FROM DOPPLER BLOOD-FLOW SIGNAL ANALYSIS WITH TIME-FREQUENCY REPRESENTATION AND PATTERN-RECOGNITION TECHNIQUES

A pattern recognition system was used to classify Doppler blood flow s ignals for the determination of lower limb arterial stenoses. The diag nostic features were extracted from time-frequency representations of Doppler signals. Three techniques were tested to estimate time-frequen cy representations: the short-time Fourier transform, the autoregressi ve (AR) modeling, and the Bessel distribution. A boundary tracking alg orithm was proposed to extract the frequency contour of the Doppler ti me-frequency representations. Based on the characteristics of the Dopp ler frequency contour, shape descriptors from an autoregressive analys is were proposed as diagnostic features. Simple algorithms were propos ed to normalize these autoregressive shape descriptors. Amplitude dist ribution of the Doppler time-frequency representation was also found u seful for stenosis classification. A total of 379 arterial segments fr om the aorta to the popliteal artery were classified by the pattern re cognition system into three categories of diameter reduction (0-19%, 2 049%, and 50-99%). The short-time Fourier transform provided an overal l accuracy of 80% (kappa = 0.38); AR modeling, 81% (kappa = 0.42); and the Bessel distribution, 82% (kappa = 0.43). All these results are be tter than those based on visual interpretation (accuracy = 76%, kappa = 0.29) performed by a trained technologist. The AR modeling and the B essel distribution improved the performance of the pattern recognition system in comparison with the short-time Fourier transform. It is lik ely that with further improvement, the pattern recognition approach wi ll be a useful clinical tool to quantify stenoses and to follow the di sease progression with more reliability and less bias than visual inte rpretation as done currently in clinical practice.