Deep learning auto-segmentation of cervical skeletal muscle for sarcopenia analysis in patients with head and neck cancer

• DOI: 10.3389/fonc.2022.930432
• Publication Date: 2022-07-28T05:29:00Z
• Authors: Mohamed A. Naser; Kareem A. Wahid; Aaron J. Grossberg; Brennan Olson; Rishab Jain; Dina El-Habashy; Cem Dede; Vivian Salama; Moamen Abobakr; Abdallah S. R. Mo...; Renjie He; Joel Jaskari; Jaakko Sahlsten; Kimmo Kaski; Clifton D. Fuller
• Keywords: Sørensen–Dice coefficient; Ground truth

Sarcopenia is a prognostic factor in patients with head and neck cancer (HNC). can be determined using the skeletal muscle index (SMI) calculated from cervical (SM) segmentations. However, SM segmentation requires manual input, which time-consuming variable. Therefore, we developed fully-automated approach to segment vertebra SM.390 HNC contrast-enhanced CT scans were utilized (300-training, 90-testing). Ground-truth single-slice segmentations at C3 manually generated. A multi-stage deep learning pipeline was developed, where 3D ResUNet auto-segmented section (33 mm window), middle slice of auto-selected, 2D auto-selected slice. Both approaches trained five sub-models (5-fold cross-validation) combined sub-model predictions on test set majority vote ensembling. Model performance primarily Dice similarity coefficient (DSC). Predicted SMI cross-sectional area. Finally, established cutoffs, performed Kaplan-Meier analysis determine associations overall survival.Mean DSC models 0.96 0.95, respectively. had high correlation ground-truth males females (r>0.96). stratified for survival (log-rank p = 0.01) but not 0.07), consistent SMI.We high-performance, multi-stage, SM. Our study an essential step towards sarcopenia-related decision-making HNC.