Deep learning-based automatic pipeline for quantitative assessment of thigh muscle morphology and fatty infiltration

Abstract

Purpose: Fast and accurate thigh muscle segmentation from MRI is important for quantitative assessment of thigh muscle morphology and composition. A novel deep learning (DL) based thigh muscle and surrounding tissues segmentation model was developed for fully automatic and reproducible cross-sectional area (CSA) and fat fraction (FF) quantification and tested in patients at 10 years after anterior cruciate ligament reconstructions.

Methods: A DL model combining UNet and DenseNet was trained and tested using manually segmented thighs from 16 patients (32 legs). Segmentation accuracy was evaluated using Dice similarity coefficients (DSC) and average symmetric surface distance (ASSD). A UNet model was trained for comparison. These segmentations were used to obtain CSA and FF quantification. Reproducibility of CSA and FF quantification was tested with scan and rescan of six healthy subjects.

Results: The proposed UNet and DenseNet had high agreement with manual segmentation (DSC >0.97, ASSD < 0.24) and improved performance compared with UNet. For hamstrings of the operated knee, the automated pipeline had largest absolute difference of 6.01% for CSA and 0.47% for FF as compared to manual segmentation. In reproducibility analysis, the average difference (absolute) in CSA quantification between scan and rescan was better for the automatic method as compared with manual segmentation (2.27% vs. 3.34%), whereas the average difference (absolute) in FF quantification were similar.

Conclusions: The proposed method exhibits excellent accuracy and reproducibility in CSA and FF quantification compared with manual segmentation and can be used in large-scale patient studies.

Keywords: automated segmentation; deep learning; magnetic resonance imaging; thigh muscle.

Figure 1:

Architecture of the proposed U-DenseNet.

Figure 2:

Overlay of segmentations from manual and automatic methods. Rows present example slices from three different subjects, and columns are segmentation provided by different methods. Arrows indicate the discrepancy between manual and auto segmentation.

Figure 3:

Bland–Altman analysis for the CSA and FF in scan-rescan data for different muscle groups. The analyses were performed to compare differences in CSA and FF quantifications between scan and rescan for six controls in different thigh muscle groups. The ±95% confidence intervals (CIs) are on top of each sub-plot. The first row figures present the analysis for CSA, and the second row figures are the analysis for FF. Each column figures presents different muscle groups. The CSA and FF measurements using the proposed method are marked by purple dots, and the CSA and FF measurements using manual segmentation are indicated by orange dots.

Figure 4:

Manual and proposed automatic segmentation overlays on scan and rescan are shown for the reproducibility analysis. The first row figures are the slice from the scan, and the second row figures are the collocated slice from rescan of the same subject. The discrepancy in manual and auto segmentation (indicated by the arrow) are shown in these figures. This subject had the highest CSA quantification difference between scan and rescan using manual segmentation.

Figure 5:

Fat fraction (FF) maps overlay using manual and automatic segmentation by proposed model. The fatty infiltration (indicated by the arrow) has replaced the muscle fully, which is causing error in automatic segmentation.