Search

Transcatheter mitral valve replacement (TMVR) has emerged as a minimally invasive alternative for treating patients suffering from mitral valve disease. The number of TMVR procedures is expected to rise as devices currently in clinical trials obtain approval for commercialization. Automating the planning of such interventions becomes, therefore, more relevant in an attempt to decrease inter-subject discrepancies and time spent in patient assessment. This study evaluates the performance of an automated method for detection of anatomical landmarks and generation of relevant measurements for device selection and positioning. Cardiac CT scans of 70 patients were collected retrospectively. Fifty scans were used to generate a statistical shape model (SSM) of the left heart chambers at ten different timepoints, whereas the remaining 20 scans were used for validation of the automated method. The clinical measurements resulting from the anatomical landmarks generated automatically were compared against the measurements obtained through the manual indication of the corresponding landmarks by three observers, during systole and diastole. The automatically generated measurements were in close agreement with the user-driven analysis, with intraclass correlation coefficients (ICC) consistently lower for the saddle-shaped (ICCArea = 0.90, ICCPerimeter 2D = 0.95, ICCPerimeter 3D = 0.93, ICCAP-Diameter = 0.71, ICCML-Diameter = 0.90) compared to the D-shaped annulus (ICCArea = 0.94, ICCPerimeter 2D = 0.96, ICCPerimeter 3D = 0.96, ICCAP-Diameter = 0.95, ICCML-Diameter = 0.92). The larger differences observed for the saddle shape suggest that the main discrepancies occur in the aorto-mitral curtain. This is supported by the fact that statistically significant differences are observed between the two annulus configurations for area (p < 0.001), 3D perimeter (p = 0.009) and AP diameter (p < 0.001), whereas errors for 2D perimeter and ML diameter remained almost constant. The mitral valve center deviated in average 2.5 mm from the user-driven position, a value comparable to the inter-observer variability. The present study suggests that accurate mitral valve assessment can be achieved with a fully automated method, what could result in more consistent and shorter pre-interventional planning of TMVR procedures.