Computer analysis and pattern recognition of doppler blood flow spectra for disease classification in the lower limb arteries

Abstract

In the present study, a computer processing method was developed to objectively classify disease in the lower limb arteries evaluated by noninvasive ultrasonic duplex scanning. This method analyzes Doppler blood flow signals, extracts diagnostic features from Doppler spectrograms and classifies the severity of the disease into three categories of diameter reduction (0–19%, 20–49% and 50–99%). The features investigated were based on frequency features obtained at peak systole, spectral broadening indices and normalized amplitudes of the power spectrogram computed in various positive and negative frequency bands. A total of 379 arterial segments studied from the aorta to the popliteal artery were classified using a pattern recognition method based on the Bayes model. Two classification schemes using a two-node decision rule were tested. Both schemes gave similar results, the first one provided an overall accuracy of 83% (Kappa = 0.42) and the second an overall accuracy of 81% (Kappa = 0.35) when compared with conventional biplane contrast arteriography. These performances, especially for the 0 to 19% lesion category, are better than the one obtained by the technologist (accuracy = 76% and Kappa = 0.33), based on visual interpretation of the Doppler spectrograms. To recognize hemodynamically significant stenoses (50–99% lesions), the pattern recognition system has a sensitivity and a specificity of 50% and 99%, respectively, using classification scheme I. With classification scheme II, the sensitivity and the specificity are 50% and 98%, respectively. Visual interpretation of the Doppler spectrograms leads to a sensitivity and a specificity of 50% and 98%, respectively. These results are the first to be obtained by a pattern recognition system in classifying lower limb arterial stenoses.