Evaluation of interrater reliability of different muscle segmentation techniques in diffusion tensor imaging

Introduction Muscle diffusion tensor imaging (mDTI) is a quantitative MRI technique that can provide information about muscular microstructure and integrity. Ultrasound and DTI studies have shown intramuscular differences, and therefore separation of different muscles for analysis is essential. The commonly used methods to assess DTI metrics in muscles are manual segmentation and tract‐based analysis. Recently methods such as volume‐based tractography have been applied to optimize muscle architecture estimation, but can also be used to assess DTI metrics. Purpose To evaluate diffusion metrics obtained using three different methods—volume‐based tractography, manual segmentation‐based analysis and tract‐based analysis—with respect to their interrater reliability and their ability to detect intramuscular variance. Materials and methods 30 volunteers underwent an MRI examination in a 3 T scanner using a 16‐channel Torso XL coil. Diffusion‐weighted images were acquired to obtain DTI metrics. These metrics were evaluated in six thigh muscles using volume‐based tractography, manual segmentation and standard tractography. All three methods were performed by two independent raters to assess interrater reliability by ICC analysis and Bland‐Altman plots. Ability to assess intramuscular variance was compared using an ANOVA with muscle as a between‐subjects factor. Results Interrater reliability for all methods was found to be excellent. The highest interrater reliability was found for volume‐based tractography (ICC ≥ 0.967). Significant differences for the factor muscle in all examined diffusion parameters were shown in muscles using all methods (main effect p < 0.001). Conclusions Diffusion data can be assessed by volume tractography, standard tractography and manual segmentation with high interrater reliability. Each method produces different results for the investigated DTI parameters. Volume‐based tractography was superior to conventional manual segmentation and tractography regarding interrater reliability and detection of intramuscular variance, while tract‐based analysis showed the lowest coefficients of variation.