Classification of arterial plaque by spectral analysis of in vitro radio frequency intravascular ultrasound data.

To test whether radio-frequency analysis of coronary plaques predicts the histological classification, r.f. data were collected using a 30 MHz intravascular ultrasound scanner. Two hundred ninety-nine regions-of-interest from eight postmortem coronary arteries were selected and identified by histology as falling into one of seven different tissue types. These are loose fibrous tissue (n = 78), moderate fibrous tissue (n = 27), dense fibrous tissue (n = 33), microcalcification (n = 14), calcified plaque (n = 55), lipid/fibrous mixture (n = 51) and homogeneous areas of lipid pool (n = 29). On the basis of a previous study, four spectral parameters were calculated for each of the regions-of-interest: maximum power (dB), mean power (dB), spectral slope (dB/MHz) over the bandwidth 18-35 MHz and the intercept of the spectral slope with the 0 Hz axis (dB). A minimum-distance classifier using the Mahalanobis (1948) distance was applied to the data. Following resubstitution of the training data into the classifier, the total correctly classified was 54%. The data were reclassified using three broader tissue groups: (1) calcified plaque, (2) lipid pool and (3) a mixed fibrous category, incorporating loose fibrous tissue, moderate fibrous tissue, dense fibrous tissue, lipid/fibrous mixture and microcalcification. The total correctly classified was 86%. Using "leave-one-out" cross-validation, the classification rates were 48% for seven tissue subgroups and 83% for three broader categories of tissue type.