Table 4.
AI-based on MRI is applied in the differential diagnosis of pancreatic cancer and other pancreatic tumors
| Reference | Sample size | Data source | Algorithms | Aim | Best result |
|---|---|---|---|---|---|
| Li et al. 146 | 267 samples from 4 modalities (T1: 67, T2: 68, DWI: 68, AP: 64) | T1, T2, DWI, AP MRI | UDA+ meta learning+ GCN | PC segmentation | DSC (62.08%, T1), (61.35%, T2), (61.88%, DWI), (60.43%, AP) |
| Chen et al. 147 | 73 cases | multi-sequences MRI | Spiral-ResUNet | PC segmentation | DSC (65.60%), Jaccard index (49.64%), HD (7.27mm), Recall (76.69%), Precision (62.96%) |
| Liang et al. 148 | 56 DCE MRI sets | DCE MRI | CNN (SGDM) | PDAC segmentation | DSC (71%), HD (7.36mm), MSD (1.78mm) |
| Goldenberg et al. 149 | 30 mouse models | T1 relaxation, CEST, and DCE MRI | SVM | PC classification | Accuracy (87.5%, CEST) (85.1%, DCE) |
| Cui et al. 150 | 202 cases | T1-w, T2-w, CET1-w MRI | LASSO | BD-IPMN grading | AUC (0.903), Specificity (94.8%), Sensitivity (73.4%) |
| Corral et al. 151 | 139 cases | multi-sequences MRI | CNN | IPMN classification | AUC (0.783), Sensitivity (75%), Specificity (78%), PPV (73%), NPV (81%) |
| Hussein et al. 56 | 171 cases | T2 MRI | SVM, RF, 3D CNN | IPMN classification | Unsupervised:Accuracy (58.04%), Sensitivity (58.61%), Specificity (41.67%); Supervised: Accuracy (84.22%), Sensitivity (97.2%), Specificity (46.5%) |
| Cheng et al. 152 | 60 cases | CE-CT, T2 MRI | LR, SVM | Malignant IPMN prediction |
MRI+SVM: AUC (0.940), Accuracy (86.7%), Sensitivity (95.7%), Specificity (81.1%), PPV (75.9%), NPV (96.8%) CT+SVM: AUC (0.864), Accuracy (83.3%), Sensitivity (78.3%), Specificity (86.5%), PPV (78.3%), NPV (86.5%) |
Abbreviations: AUC: area under the curve; BD-IPMN: branching type IPMN; CEST: chemical exchange saturation transfer; CNN: convolutional neural network; DCE: dynamic contrast enhancement; DSC: dice similarity coefficient; GCN: Graph Convolutional Networks; HD: Hausdorff distance; IPMN: intraductal papillary mucinous neoplasm; LR: logistic regression; MLP: multilayer perceptron; MSD: mean surface distance; NPV: negative predictive value; PC: pancreatic cancer; PDAC: pancreatic ductal adenocarcinoma; PPV: positive predictive value; RF: random forest; SGDM: stochastic gradient descent with momentum; SVM: support vector machine; UDA: unsupervised domain adaptation; AP MRI: atrial phase MRI; DWI: diffusion weighted imaging.