Development of an initial training and evaluation programme for manual lower limb muscle MRI segmentation

Area of Science:

• Neuromuscular Diseases
• Medical Imaging
• Biomarker Development

Background:

• Magnetic resonance imaging (MRI) quantification of intramuscular fat is a key biomarker for neuromuscular diseases.
• Manual muscle segmentation is foundational for automated methods.
• Reliability benchmarks are needed for human and machine learning segmentation.

Purpose of the Study:

• Develop a training program for lower limb muscle segmentation in MRI.
• Demonstrate competence in manual segmentation for new observers.
• Establish reliability benchmarks for future segmentation efforts.

Main Methods:

• A training program included a manual, direct instruction, and eight lower limb MRI scans.
• Assessment used test-retest scans with interscan and interobserver reliability metrics.
• Competency benchmarks were determined using Sørensen-Dice similarity coefficients.

Main Results:

• Six of eight new observers achieved reliability comparable to experienced observers.
• Benchmarks established: Sørensen-Dice >0.87 (thigh) and >0.92 (calf) for individual muscles.
• Large regions of interest (ROIs) demonstrated significantly higher reliability than small ROIs.

Conclusions:

• The first formal training program for manual lower limb muscle segmentation was developed and analyzed.
• Large ROIs are more reliable for fat fraction assessment than small ROIs.
• Established competency benchmarks are crucial for quantitative muscle MRI biomarkers in neuromuscular diseases.