Evaluation study of fast spectral estimators using in-vivo data

Spectrograms in medical ultrasound are usually estimated with Welch's method (WM). To achieve sufficient spectral resolution and contrast, WM uses an observation window (OW) of up to 256 emissions per estimate. Two adaptive filterbank methods have been suggested to reduce the OW: Blood spectral Power Capon (BPC) and the Blood Amplitude and Phase EStimation method (BAPES). Ten volunteers were scanned over the carotid artery. From each dataset, 28 spectrograms were produced by combining four approaches (WM with a Hanning window (W.HAN), WM with a boxcar window (W.BOX), BPC and BAPES) and seven OWs (128, 64, 32, 16, 8, 4, 2). The full-width-at-half-maximum (FWHM) and the ratio between main and side lobe levels were calculated for each spectrogram. Furthermore, all 280 randomized spectrograms were presented to nine radiologists for visual evaluation: useful/not useful. BAPES and BPC compared to WM had better resolution (lower FWHM) for all OW<128 while only BAPES compared to WM had improved contrast (higher ratio). According to the scores given by the radiologists, BAPES, BPC and W.HAN performed equally well (p>0.05) at OW 128 and 64, while W.BOX scored less (p<0.05). At OW 32, BAPES and BPC performed better than WM (p<0.0001) and BAPES was significantly superior to BPC at OW 16 (p=0.0002) and 8 (p<0.0001). BPC at OW 32 (p=0.29) and BAPES at OW 16 (p=0.55) scored as at OW 128. WM at OW 16 and 8 failed as the four methods at OW 4 and 2. The intra-observer variability tested for three radiologist showed on average good agreement (90%, ?=0.79) and inter-observer variability showed moderate agreement (78%, ?=0.56). The overall result indicates that BPC and BAPES have better resolution and BAPES better contrast than WM, and that OW can be reduced to 32 using BPC and 16 using BAPES without reducing the usefulness of the spectrogram. This could potentially increase the temporal resolution of the spectrogram or the frame rate of the interleaved B-mode images.