Skeletal muscle segmentation from MRI dataset using a model-based approach

Abstract

Magnetic resonance imaging (MRI) and computed tomography scans are used to assess muscle volume, but the manual segmentation, slice by slice, is long and tedious. We proposed an improvement in the deformation of a parametric-specific object method using image processing. The 3D subject-specific geometry was reconstructed based on a few selected number of MRI slices by fast rough contouring using polygons. These polygons were matched to the muscle shape by an optimisation method using an original cost function. Then, parametric-specific object was constructed and deformed. The shape was improved using a loop and the cost function in all MRI slices. The 11 main muscles of the thigh were considered, and the time required to get the shape of all muscles was 21 min, with a volume error inferior to 5% and a point-surface distance error (2RMS) inferior to 5 mm. This method provides a good compromise between segmentation time and an accurate representation of the muscles shape.