Evaluation of an automated real-time spectral analysis technique

An adaptive real-time Doppler peak-frequency tracing algorithm was evaluated in vitro and compared to manual peak-frequency traces. A computer-controlled pump was used to generate physiological flow waveforms in a vasculature-mimicking phantom. Spectral waveforms were obtained on an ATL HDI along with real-time estimates of diagnostic parameters, including maximum systolic, minimum diastolic, time-averaged peak frequencies and pulsatility and resistance indices. The effect of the signal-to-noise ratio on the measured parameters was investigated. The imprecision in the measured parameters was found to depend somewhat on the waveform shape; e.g., the imprecision in PI was 4.1% for a normal renal waveform and 8.5% for a waveform having reverse diastolic flow. The peak frequency envelopes of the same waveform data were traced manually by nine operators, and the resulting diagnostic parameters were compared to ones obtained from automated peak-frequency traces of the same waveform data. The agreement between parameters measured by the automated routine and those measured manually was found to depend somewhat on the waveform shape; e.g., the bias in the PI was 1.3% for a renal waveform lacking diastolic flow, and 12% for a waveform with reverse diastolic flow. The between-observer variations in the manual measurements ranged from 0.8% up to 9.4%. The overall variations associated with the automated traces were found to be smaller than or equal to those of the manual traces.