Assessment of regional myocardial function using 3D cardiac strain estimation: comparison against conventional echocardiographic assessment

Our labs introduced methods that are able to assess regional myocardial function, from volumetric ultrasound data. Burdock estimates myocardial deformation using a 3D speckle-tracking approach, while splineMIRIT is a B-spline transformation-based elastic registration method. The aim of the present study was to test and contrast the diagnostic accuracy of both approaches in the clinical setting. Furthermore, the manual initialization step, required by both methods, was replaced by using RCTL, a real time tracking, fully automatic library that provided the initial geometry of the left ventricle. Volumetric ultrasound data was acquired in a total of 12 healthy volunteers and 13 patients with proven coronary artery disease. The acquired data was divided in two subgroups. For the first, wall motion scoring (WMS) (0=normo-kinetic; 1=hypo-kinetic; 2=a-kinetic; 3=dys-kinetic), 3D image quality (0 = good, 1 = reasonable, 2 = poor, 3= unacceptable) as well as tracking quality (0=bad; 1=good) using three standard apical views and 3 short axis slices at different levels in the LV were recorded on a segmental basis by a trained cardiologist blinded to the data and compared among methods. Their ability to detect infarcted patients was tested by computing per patient sensitivity and specificity values (WMS as reference). For the second group segmental longitudinal strain values were computed using a commercially available package as well as Burdock and splineMIRIT. Differences in normal and unhealthy segments of high 3D quality were compared. Tracking quality was better for splineMIRIT (85% vs. 79% of segments), however Burdock achieved better detection of MI cases (sensitivity 70% vs. 40%, specificity 81.82% vs. 90.91%). In addition, the ANOVA test yielded significant differences, in segmental longitudinal stain, among normal and abnormal segments for both methods (p-value < 0.01). The presented methods are fully automatic and results show that they can distinguish between no- rmal and unhealthy segments in clinical recordings.