Assessment of manual segmentation of magnetic resonance images of skeletal muscles

Technical considerations related to the manual detection of muscles boundaries from magnetic resonance images (MRI) are evaluated. Two commercial image processing software programs were used by two operators to obtain measurements on MRI from a phantom and from muscles of the upper limb and trunk. Optimization of MRI acquisition sequences, image resolution, image contrast, and sub-sampling effect were also experimented. No significant intra- and interoperator variation was observed and results obtained from both softwares were similar. Generally, sub-millimetric slice thickness offers better definition but lower contrast than thicker slices. Differences in estimated length, surface and volume of upper limb and back muscles were small for slice thickness varying between 1.5 and 4 mm. Accuracy of manual segmentation of muscles with MRI was found more dependent on the contrast than on the human factors.