Table 1.
Studies on segmentation of white matter lesion changes.
| Study | n | MRI parameters, voxel size (mm) | Description/comments | Lesion-wise performance parameters (range)a |
|
|---|---|---|---|---|---|
| TPR | FDR/FPR | ||||
| (Elliott et al., 2010) | 23 | 1.5 T; T1w, PD/T2w: 1 × 1 × 3 | Bayesian classification framework on subtraction images (T2w), training dataset (n = 66); specificity not quantitatively evaluated | 0.84 (n.i.) | n.d. |
| (Sweeney et al., 2013) | 5 | 1.5 T; 2D FLAIR, PD, T2w, 3D T1w: 1 × 1 × 1 (extrapolated) | Logistic regression model using multiple sequences and subtraction images, ROC curve analysis | 0.95 (voxel-wise) | 0.01 (voxel-wise) |
| (Battaglini et al., 2014) | 19 | PD, T2w, T1w, FLAIR: 3 × 1 × 1 | Based on subtraction images (PD); multicenter trial (Miller et al., 2012) | 0.91 (n.i.) | 0.21 (n.i.) |
| (Ganiler et al., 2014) | 20 | 1.5 T; PD, T2w, T1w: 3 × 1 × 1 | Based on subtraction images (PD) | 0.77 (n.i.) | 0.18 (n.i.) |
| (Cabezas et al., 2016) | 36 | 3 T; PD/T2w: 0.8 × 0.8 × 3; FLAIR: 0.5 × 0.5 × 3 T1w: 1 × 1 × 1.2 | Multichannel pipeline based on deformation fields | 0.81 (n.i.) | 0.18 (n.i.) |
| (Jain et al., 2016) | 12 | 3 T; T1w, FLAIR: 1 × 1 × 1 | Expectation-maximization framework | 0.62 (0.53–0.69) | 0.16 (0.00–0.51) |
| (Eichinger et al., 2017) | 106 | 3 T; FLAIR: 1 × 1 × 1.5; T1w: 1 × 1× 1 | Based on subtraction images (FLAIR); relating to a consensus reference, main focus on DIR subtraction images | 0.79 (n.i.) | 0.05(n.i.) |
| (patient-wise) | (patient-wise) | ||||
| (Salem et al., 2018) | 60 | 3 T; PD/T2w: 0.8 × 0.8 × 3; FLAIR: 0.5 × 0.5 × 3; T1w: 1× 1× 1.2 | Multichannel pipeline, use of intensities, subtraction images, and deformation fields; 36 MS patients, 24 controls | 0.74 (+/− 0.29) | 0.12 (±0.18) |
a
Performance parameters (with ranges as given in the original publications) refer to lesions unless indicated by italic letters; FLAIR, fluid attenuated inversion recovery; FDR/FPR, false discovery/positive rate; MS, multiple sclerosis; MRI, magnetic resonance imaging; n.d., not determined; n.i., not indicated; PD, proton density; ROC, response operator characteristics; T, Tesla; TPR, true positive rate, i.e. detection rate or sensitivity; w, weighted.