Table 3.
Model performance of conventional machine learning algorithms.
| Clinical outcome | Imaging modality | Clinical variable | Model | References | Sample size (T/EV) | Model performance | Validation | ||
|---|---|---|---|---|---|---|---|---|---|
| AUC (95% CI) | Others | Internal | External | ||||||
| Good functional outcome at 90 days (mRS ≤ 2 or mRS ≤ 3) | NCCT and CTA | Yes | Gradient boosting decision trees | Brugnara et al. (22) | 246 | 0.74 (0.73–0.75) | ACC,0.71 | Yes | No |
| Yes | RFA | Van et al. (23) | 1,383* | 0.79 (0.79–0.79) | n.a. | Yes | No | ||
| NCCT | Yes | Regression trees | Alawieh et al. (24) | 110/36 | Internal: 0.93 (0.85–1.00)† External: n.a. |
Internal: n.a. External: PV+: 0.60, NV-: 0.95 | Yes | Yes | |
| Yes | RLR | Nishi et al. (25) | 387/115 | Internal: 0.86 (0.78–0.94)† External:0.90 (0.83–0.97)† |
Internal: ACC,0.75; SEN,0.59; SPE,0.86; External: n.a. | Yes | Yes | ||
| MRI (DWI and PWI) | Yes | RFA | Hamann et al. (26) | 222 | 0.68 (0.61–0.76) | n.a. | Yes | No | |
| Yes | RFA | Kerleroux et al. (27) | 133 | 0.83 (0.74–0.92)† | ACC,0.73; SEN,0.69; SPE,0.76 | Yes | No | ||
| MRI(DWI) | Yes | SVM | Xie et al. (28) | 143 | 0.82 (0.75–0.89)† | ACC,0.77 | Yes | No | |
| Poor functional outcome at 90 days (mRS≥5 or mRS≥4) | NCCT and CTA | Yes | ANN | Ramos et al. (29) | 1,401* | 0.81 (0.79–0.83) | ACC, 0.65; SEN, 0.53; SPE,0.89; PV+, 0.69; NV-,0.80 | Yes | No |
| CTA | Yes | SVM | Ryu et al. (30) | 482 (hold-out testing: 208) | 0.82 (0.76–0.87) | n.a. | Yes | No | |
| n.a. | Yes | Decision trees | Kappelhof et al. (31) | 1,090* | n.a | ACC,0.72 | Yes | No | |
| Successful reperfusion (TICI score≥2b) | NCCT and CTA | Yes | RFA | Van et al. (23) | 1,383* | 0.55 (0.55–0.56) | n.a. | Yes | No |
| Successful reperfusion at the first attempt (TICI score≥2b) | NCCT and CTA | No | RLR | Patel et al. (32) | 119 | 0.77 (0.54–0.90) | ACC, 0.74 | Yes | No |
| NCCT and CTA | No | SVM | Hofmeister et al. (33) | 109/47 | External: 0.88 (0.75–1.00)† | External: ACC, 0.85; SEN, 0.50; SPE, 0.97, PV+, 0.86; NV-,0.85 | Yes | Yes | |
ANN, artificial neural networks; AUC, area under the Receiver Operating Characteristic curve; ACC, accuracy; CTA, computed tomography angiography; DWI, diffusion weighted imaging; EV, external validation dataset; MRI, magnetic resonance imaging; NCCT, non-contrast computed tomography; NV-, negative predictive value; PWI, perfusion weighted imaging; PV+, positive predictive value; RFA, random forest analysis; RLR, regularized logistic regression; SVM, support vector machine; SEN, sensitivity; SPE, specificity; T, training dataset; n.a., not available/not applicable. Note:
model derived from patients registered in MR CLEAN Registry (38).
95% CI was estimated based on normal distribution.