A robust model-based approach to detect the mitral annulus in 3D ultrasound

Over 40.000 mitral reconstructions are performed each year in the United States. To ensure a successful and durable outcome of the operation, detailed quantification of the mitral annulus is helpful. However, manual measurement is time consuming and hard to perform in clinical routine. We propose a fast semi-automatic method to create a precise model of the mitral annulus from 3D ultrasound data. The basic idea is to combine image information with anatomical knowledge in form of a standard mitral annulus model. This way, the method can adjust to the individual image data and still cope with strong artifacts and incomplete images. By comparing the resulting models to manually created ground truth data of 39 patients, we identified a mean error of 3.49 mm. This is lower than the determined standard deviation of the expert (4.13 mm) and confirms the accuracy of the proposed method. The overall time to create a mitral annulus model from 3D ultrasound image data is less than a minute. Due to its speed, accuracy and robustness, the method is eligible for the clinical routine.