Optimization of a multi-atlas method for automated segmentation of the gluteal muscles from MRI

SARMIENTO, FLORENCIA ; BELZUNCE, MARTIN A.

San Juan

Congreso; XXIII CONGRESO ARGENTINO DE BIOINGENIERÍA Y XII JORNADAS DE INGENIERÍA CLÍNICA; 2022

SABI

The assessment of the gluteal muscles is of interest in a wide number of applications. MRI can provide measures of muscle size and fat infiltration, however the image segmentation of each individual muscle is needed to compute these metrics. Multi-atlas segmentation has proven to have good results in this regard. The purpose of this work is to optimize a multi-atlas method for the automated segmentation of the gluteal muscles from magnetic resonance images, by assessing different labels fusion strategies and configurations. To achieve this, we compared four different label fusion strategies working with 1 to 23 selected label images in the fusion process. Using a leave one out strategy, we measured the Dice similarity coefficients for the segmentation of 24 images for each of the fusion methods. The optimal number of selected images to fuse was 5 and 6, depending on the method. After the optimization of the number of selected images, there were no relevant differences in the segmentation performance. This would indicate that simple fusion strategies, such as majority voting, would be appropriate if the number of selected images is optimized.