Development and validation of a deep learning model for liver shear stiffness regression using abdominal multiparametric MRI across multiple sites and vendors.
Authors
Affiliations (13)
Affiliations (13)
- Imaging Research Center, Department of Radiology, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, USA.
- Department of Radiology, University of Cincinnati College of Medicine, Cincinnati, OH, USA.
- Neurodevelopmental Disorders Prevention Center, Perinatal Institute, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, USA.
- Artificial Intelligence Imaging Research Center, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, USA.
- Department of Radiology, University of Wisconsin, Madison, WI, USA.
- Departments of Medical Physics, Biomedical Engineering, Medicine, and Emergency Medicine, University of Wisconsin, Madison, WI, USA.
- Department of Radiology, Michigan Medicine, Ann Arbor, MI, USA.
- Department of Radiology, New York University, New York, NY, USA.
- Department of Pediatrics, University of Cincinnati College of Medicine, Cincinnati, OH, USA.
- Department of Computer Science, Biomedical Engineering, Biomedical Informatics, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, USA.
- Imaging Research Center, Department of Radiology, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, USA. [email protected].
- Department of Radiology, University of Cincinnati College of Medicine, Cincinnati, OH, USA. [email protected].
- Artificial Intelligence Imaging Research Center, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, USA. [email protected].
Abstract
Liver stiffness measurement is important for assessing chronic liver disease (CLD). MR elastography (MRE) requires specialized hardware and expertise. Non-invasive deep learning (DL) models using multiparametric abdominal MRI may provide an accessible alternative. We sought to develop and validate a DL model for predicting continuous liver shear stiffness from non-contrast multiparametric abdominal MRI and electronic health record (EHR) data across multiple sites and vendors. This was a retrospective, multi-institutional study. We analyzed 3680 abdominal MRI examinations from 3376 patients with confirmed or suspected CLD. Non-contrast T1-weighted (T1w), T2-weighted (T2w), and diffusion-weighted imaging (DWI) with EHR data were used as inputs. MRE-derived liver shear stiffness served as the reference. A transformer-based multi-channel DL model was trained using multi-site 10-fold cross-validation and evaluated on temporally held-out internal (n = 1224) and independent external (n = 365) test sets. Performance was measured by Pearson's correlation coefficient (r); residual analysis assessed bias. In cross-validation, the model achieved an r of 0.78 (95% CI: 0.75, 0.80). On the internal test set, r was 0.77 (95% CI: 0.73, 0.80), and on the external set, r was 0.76 (95% CI: 0.69, 0.83). The model showed no significant bias based on age, sex, or BMI (p > 0.05). In patients with and without steatotic liver disease, r was 0.74 and 0.76, respectively. Our transformer-based multi-channel model predicts continuous liver shear stiffness from routinely acquired multiparametric MRI and EHR data with moderate correlation to MRE, representing a potential step toward accessible, non-invasive liver stiffness estimation. Question Can routinely acquired multiparametric abdominal MRI and electronic health record data predict liver stiffness across multiple sites and scanner vendors using a deep learning approach? Findings The optimized deep learning model predicted liver stiffness with r = 0.78 in cross-validation and r = 0.76 in external validation using multiparametric MRI and electronic health record data. Clinical relevance This study introduces a preliminary yet robust AI method to estimate liver stiffness from routine multiparametric MRI and EHR data, offering a scalable fibrosis assessment approach suitable for opportunistic evaluation and as a complementary tool when MRE is unavailable.