Optimisation of the automated segmentation of the mitral valve plane in 4D flow cardiovascular MRI

Abstract

Mitral valve segmentation is an important procedure to help evaluate blood flow in mitral regurgitation. Four-dimensional cardiovascular magnetic resonance flow imaging (4D flow MRI) is a technique that provides useful information for the quantification and accurate assessment of flow. This study proposes an algorithm that utilises a minimum cost function for segmenting the mitral valve area using CMR cine images and 4D flow MRI images of 8 subjects. The minimum cost function based algorithm is used to segment mitral opening at the time frames with mitral flow above a threshold (60\% of the maximum flow). In the time frames with high flow, hyper-parameter optimisation is performed and the minimum cost function is applied iteratively to increase accuracy. At time frames with low to no mitral flow, the mitral opening is segmented by interpolating the closest minimum cost function based contours. Qualitatively, the algorithm resulted in accurately localised contours, and the mitral valve area was outlined for all time frames. The algorithm's performance on the time frames with high flow is evaluated by comparing them with manually drawn (ground truth) contours. Quantitatively, 72\% of data time frames tested showed improved DSC before and after parameter optimisation. The application of the Wilcoxon Signed-Rank Test revealed no statistically significant improvements between the DSC for the whole study (W = 527, p = 0.444). The method has been implemented in the QFlow4D application of Medis. Future work includes testing the algorithm on a larger database, registering cine CMR images and 4D flow MRI images and improving the contour at the time frames with low flow by excluding the aorta from the interpolated mitral valve segmentation.