Technical details of machine learning algorithms (10 studies)
| Extracting ROI | Classification models | Methods of evaluating performance | Ground truth annotations | |
| Bordner et al21 | Automatically | DL: Mask-RCNN | Internal and external test | ASAS MRI sacroiliitis±of each persons’ MRI |
| Ye et al22 | Manually (two MS radiologists) | mRMR and LASSO | Validation | ASAS clinical diagnosis of AxSpA±of each person |
| Tenório et al23 | Manually (two MS radiologists) | ANN | Validation | ASAS MRI sacroiliitis±of each persons’ MRI |
| Roels et al24 | Automatically | DL: ResNet18 | Validation and external test | BMO±of each persons’ MRI |
| Lin et al25 | Automatically | DL: CNN (+ Attention-U-Net) | Internal test | ASAS MRI sacroiliitis±of each persons’ MRI or each images |
| Bressem et al26 | Not applicable* | DL: Res-Neural network | Validation and external test | ASAS MRI sacroiliitis ±, structural changes of SI joints ±, active inflammatory changes of SI joints±of each persons’ MRI |
| Nicolaes et al27 | Manually (three MS radiologists) | DL: unclear architecture | Validation | BMO±of each persons’ MRI |
| Lee et al28 | Manually (two rheumatologists and one radiologist) | DL: CNN (+ResNet18) | Validation | ASAS MRI sacroiliitis±of each persons’ MRI or each images |
| Kepp et al29 | Manually (two radiologists) | ANN (+k-nearest neighbour) | Validation | ASAS MRI sacroiliitis±, TIRM positive ASAS sacroiliitis±, MRI sacroiliitis versus degenerative changes of each persons’ MRI |
| Faleiros et al30 | Manually (one MS radiologist) | ANN, SVM, k-nearest neighbour | Validation and internal test | ASAS MRI sacroiliitis±of each persons’ MRI |
*Not extracted ROI.
ANN, artificial neural network; ASAS, Assessment of Spondyloarthritis International Society; AxSpA, axial spondyloarthritis; BMO, bone marrow oedema; CNN, convolutional neural network; DL, deep learning; LASSO, least absolute shrinkage and selection operator; mRMR, minimum-redundancy-maximum-relevance; MS, muscular skeleton; RCNN, regions with convolutional neural networks; ROI, region of interest; SI, sacroiliac; SVM, support vector machine; TIRM, Turbo Inversion Recovery Magnitude.