Deep learning-based classification of erosion, synovitis and osteitis in hand MRI of patients with inflammatory arthritis
Authors' conclusion
We developed a CNN-based automated scoring system that allowed a rapid grading of erosions, osteitis and synovitis with good diagnostic accuracy and using less MRI sequences compared with conventional scoring.
Study context
- Population
- Rheumatoid arthritis (RA) and psoriatic arthritis (PsA) patients with hand MRI, rheumatology department of Erlangen University Hospital; external validation in a second patient cohort112 patients training
- Comparison
- Convolutional neural network (CNN)-based automated scoring vs expert rheumatologist annotation
- Outcome
- Macro-AUC and balanced accuracy of automated scoring of erosions, osteitis and synovitis vs human annotations
Methodological detailsconvolutional neural network, fivefold cross-validation, external validation, 211 MRIs from 112 patients (14 906 region of interests (ROIs)) for training/internal validation; 220 MRIs from 75 patients (11 040 ROIs) for external validation
- Convolutional neural network
- Fivefold cross-validation
- External validation
- 211 MRIs from 112 patients (14 906 ROIs) for training/internal validation; 220 MRIs from 75 patients (11 040 ROIs) for external validation
Performance
As reported in the abstract
Macro-AUC, mean (SD)
| Test or model | macro-AUC erosions | macro-AUC osteitis | macro-AUC synovitis |
|---|---|---|---|
| CNN | 92%±1% | 91%±2% | 85%±2% |
CNN
macro-AUC erosions92%±1%
macro-AUC osteitis91%±2%
macro-AUC synovitis85%±2%
Further findings
| Outcome | CNN |
|---|---|
| Spearman correlation of CNN with human annotation, mean | |
| Erosions | 90±2% |
| Osteitis | 78±8% |
| Synovitis | 69±7% |
Spearman correlation of CNN with human annotation, mean
- Remained consistent in the validation dataset
Original publication
Deep learning-based classification of erosion, synovitis and osteitis in hand MRI of patients with inflammatory arthritis. RMD Open. Published 17 June 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →