Knowledge-based method for quantification of Doppler spectral data

Summary form only given. A knowledge-based image processing method for automatic spectral boundary detection in Doppler images has been developed. Unlike boundary-detection algorithms which identify spectral edges as regions of rapid changes in gray level, the method utilizes high-level contextual knowledge. The lower level image analysis involves area-of-interest preprocessing, directional-gradient-operator computation, and an edge-point linking scheme. The higher level portion incorporates information about typical spectral densities of continuous-wave (CW) and pulse-wave (PW) Doppler envelopes, envelope geometry, and normal ranges of blood flow velocities and pressure gradients for all measurement sites. Eleven normal and fourteen abnormal Doppler recordings were acquired online from aortic, mitral, tricuspid, pulmonary, and aortic sites using the Dextra D-200 image analysis system. Measurements of peak and mean pressure gradients and velocities from automatically detected spectral boundaries were compared with computer-derived measurements from manually derived boundaries by three independent observers and the accuracy of the automatic approach