Comparison of spectral analysis techniques for computer assisted classification of Doppler ultrasound spectra recorded in the lower limb arteries

The diagnostic performance of two spectral techniques (the FFT and the AR modelling) to compute Doppler spectrograms and two methods (spectrum-by-spectrum threshold and edge detection by image processing) to estimate the frequency contours of the Doppler spectrograms was evaluated in 379 Doppler blood flow signals recorded in lower limb arteries. Sixteen diagnostic features were extracted from each spectrogram and a pattern recognition system based on a two-node decision tree was used to determine the performance of each method to classify correctly arterial lesions into three classes: 0–19%, 20–49% and 50–99% diameter reduction. Results show that the performance of the classifier is more sensitive to the frequency contour algorithm than the spectral technique. AR modelling and image processing provided the best results to correctly classify the 50–99% lesions while the FFT and image processing provided the best overall percentage of correct classifications.