Novel stochastic framework for automatic segmentation of human thigh MRI volumes and its applications in spinal cord injured individuals

This study introduces a new automated method to measure muscle and fat in thigh MRI scans, particularly for individuals with spinal cord injuries. The method aims to improve accuracy and speed compared to manual measurements. The approach uses a combination of statistical techniques and image analysis to identify and segment different types of tissue, including subcutaneous fat, inter-muscular fat, and various muscle groups. The method's accuracy was tested on both individuals with spinal cord injuries and those without, and its performance was compared against existing segmentation techniques.

• 1
  The proposed automatic segmentation method achieved an overall accuracy of 0.93±0.06 for adipose tissue and muscle compartments segmentation based on Dice Similarity Coefficient.
• 2
  The novel framework demonstrated higher accuracy in muscle compartment segmentation compared to ANTs and STAPLE, which are two validated atlas-based segmentation methods.
• 3
  The framework showed comparable Dice accuracy and superior Hausdorff distance measure compared to DeepMedic, a convolutional neural network for 3-D medical image segmentation.