Qualitative and quantitative accuracy of ultrasound-based virtual histology for detection of necrotic core in human coronary arteries.

The aim was to test the accuracy of virtual histology (VH) for detecting necrotic core (NC) in an ex vivo human model of coronary arteries as compared to real histology (RH). A total of nine consecutive explanted hearts were included in the study. Coronary segments, clearly identified by anatomical landmarks, were analyzed by intravascular ultrasound (IVUS-VH) immediately after heart collection and thereafter by RH. NC was expressed as absolute (total amount) and relative (corrected for plaque area) values. Correlation analysis was performed using linear regression models at cross-section level, with correction for repeated measurements per patient, and at segment level. Receiver operator curves (ROC) were developed for testing accuracy of VH in detecting RH-NC. Overall, 321 mm were analyzed corresponding to 642 IVUS-VH frames and corresponding histological slices. VH and RH-NC areas were 0.24 ± 0.43 and 0.16 ± 0.43 mm(2), respectively (p < 0.001). At cross-section level, the correlation between VH and RH-NC was moderate in absolute (r = 0.50, p < 0.001) and poor in relative values (r = 0.43, p = 0.120). At the segment level, this correlation improves in terms of absolute values (r = 0.80, p = 0.01), but was not significant in terms of relative values (r = 0.43, p = 0.25). The ROC curve showed a C-statistics of 0.904 (p < 0.001) with high sensitivity (94 %), but low specificity (53 %) and low positive predictive value (48 %). Although VH has a high sensitivity in identifying RH-NC, it has a low specificity and low positive predictive value. In addition, it is not able to accurately quantify its size within the corresponding histological specimen.