Machine learning assisted prediction of cartilage histology using MR fingerprinting - a preliminary study.
Authors
Affiliations (8)
Affiliations (8)
- Research Unit of Health Sciences and Technology, University of Oulu, Oulu, Finland. [email protected].
- Research Unit of Health Sciences and Technology, University of Oulu, Oulu, Finland.
- Department of Technical Physics, University of Eastern Finland, Kuopio, Finland.
- Department of Diagnostic Radiology, Oulu University Hospital, Oulu, Finland.
- Medical Research Center, University of Oulu and Oulu University Hospital, Oulu, Finland.
- Donders Centre for Cognitive Neuroimaging, Donders Institute for Brain, Cognition and Behaviour, Radboud University, Nijmegen, the Netherlands.
- Bernard and Irene Schwartz Center for Biomedical Imaging, Department of Radiology, New York University Grossman School of Medicine, New York, NY, USA.
- Department of Radiology, Center for Advanced Imaging Innovation and Research (CAI2R), New York University Grossman School of Medicine, New York, NY, USA.
Abstract
Prediction of cartilage structural properties through MRI could allow earlier detection of joint pathologies, such as osteoarthritis. Bovine patellar cartilage samples (n = 12) were imaged using magnetic resonance fingerprinting, followed by histological examination of proteoglycan content and collagen fiber anisotropy. The relaxation time maps and raw signal data were then used for training Gaussian process regression (GPR) models to predict the histology results. Proteoglycan content was predicted by the GPR models with high accuracy (median r = 0.81, σ = 0.08 and NRMSE = 11.7%). Predictions performed using raw MRF data outperformed those done using qMRI maps. Collagen fiber anisotropy predictions found only weak correlation (median r = 0.40, σ = 0.25 & NRMSE = 26.4%) and no significant difference was seen between models trained on raw MRF or relaxation time maps. These findings indicate that noninvasive prediction of proteoglycan content in cartilage from MRF measurements using a 3 T clinical scanner is feasible, holding promise for future clinical applications. Collagen fiber anisotropy could not be reliably estimated with the current setup. GPR-based prediction models were found to outperform reference linear models using the same prediction data.