3D Segmentation of Pathological Muscle with a Physics-Informed Latent-Regularized U-Net
- 1UNC Chapel Hill
Abstract
ObjectiveWe developed a data-efficient deep learning framework for three-dimensional segmentation of pathological musculoskeletal anatomy from magnetic resonance imaging (MRI) when only limited manual annotations are available.
MethodsWe developed a Physics-Informed Latent-Regularized U-Net (PILR-U-Net) that combines transfer learning from healthy MRI, latent-space anatomical regularization, and elasticity-based physics-informed constraints. The physics-informed loss enforces mechanical equilibrium, near-incompressibility, and spatial smoothness of predicted deformation fields. The framework was evaluated on MRI datasets from 50 participants with cerebral palsy across 15 lower-limb musculoskeletal structures using sparse manual annotations. Performance was assessed using volumetric overlap, boundary accuracy, volume error, sensitivity, and precision. Ablation experiments evaluated the individual contributions of latent-space and physics-informed regularization.
ResultsOur experimental results show accurate segmentations with three-dimensional Dice coefficients ranging from 0.750 to 0.943 across evaluated structures, while most structures exhibited low surface-distance errors. Predicted deformation fields maintained positive Jacobian determinants near unity and smooth strain-energy distributions. Ablation analysis showed that both regularization components improved performance, with removal of physics-informed regularization producing the largest reductions in Dice accuracy and increases in boundary error.
ConclusionPILR-U-Net enables accurate and anatomically plausible segmentation of pathological musculoskeletal MRI under sparse supervision.
SignificanceIncorporating anatomical priors and biomechanical constraints into deep segmentation networks may reduce dependence on extensive pathological annotations and support patient-specific musculoskeletal modeling and clinical analysis.
Related
- TiLCA-Net: nuclei segmentation and classification in pathological images via task-specialized attention and efficient decoding
- Biologically informed network-regularized sparse consensus clustering (BioNet-SPACC) identifies a continuous plasma proteomic axis in major depressive disorder
- Automated thigh muscle segmentation using foundation models: validation and impact of segmentation quality
- Quantitative super-resolution imaging with deep learning segmentation reveals early mitochondrial network remodeling and altered mitochondria-lysosome interactions in Pompe disease muscle
- Benchmarking open-source automated thigh muscle MRI segmentation algorithms
BibTeX
@misc{mehrabi2026segmentation,
title = {{3D Segmentation of Pathological Muscle with a Physics-Informed Latent-Regularized U-Net}},
author = {N. Mehrabi and N. C. R. Pegard and G. G. Handsfield},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.15.26363155},
url = {https://www.medrxiv.org/content/10.64898/2026.09.15.26363155v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - 3D Segmentation of Pathological Muscle with a Physics-Informed Latent-Regularized U-Net
AU - N. Mehrabi
AU - N. C. R. Pegard
AU - G. G. Handsfield
PY - 2026
DA - 2026/09/15
PB - medRxiv
DO - 10.64898/2026.09.15.26363155
UR - https://www.medrxiv.org/content/10.64898/2026.09.15.26363155v1
AB - ObjectiveWe developed a data-efficient deep learning framework for three-dimensional segmentation of pathological musculoskeletal anatomy from magnetic resonance imaging (MRI) when only limited manual annotations are available. MethodsWe developed a Physics-Informed Latent-Regularized U-Net (PILR-U-Net) that combines transfer learning from healthy MRI, latent-space anatomical regularization, and elasticity-based physics-informed constraints. The physics-informed loss enforces mechanical equilibrium, near-incompressibility, and spatial smoothness of predicted deformation fields. The framework was evaluated on MRI datasets from 50 participants with cerebral palsy across 15 lower-limb musculoskeletal structures using sparse manual annotations. Performance was assessed using volumetric overlap, boundary accuracy, volume error, sensitivity, and precision. Ablation experiments evaluated the individual contributions of latent-space and physics-informed regularization. ResultsOur experimental results show accurate segmentations with three-dimensional Dice coefficients ranging from 0.750 to 0.943 across evaluated structures, while most structures exhibited low surface-distance errors. Predicted deformation fields maintained positive Jacobian determinants near unity and smooth strain-energy distributions. Ablation analysis showed that both regularization components improved performance, with removal of physics-informed regularization producing the largest reductions in Dice accuracy and increases in boundary error. ConclusionPILR-U-Net enables accurate and anatomically plausible segmentation of pathological musculoskeletal MRI under sparse supervision. SignificanceIncorporating anatomical priors and biomechanical constraints into deep segmentation networks may reduce dependence on extensive pathological annotations and support patient-specific musculoskeletal modeling and clinical analysis.
ER -