Table 1.
Radiomics studies in diagnosing precancerous pancreatic diseases.
| Reference | Image | Software | Endpoints | Segmentation Process (Number of Readers) | Sample Size (Training + Validation) | Number of Features Extracted | Results |
|---|---|---|---|---|---|---|---|
| Attiyeh et al. [36] |
CT | In-house software in MATLAB | BD-IPMN risk | manual (1) | 103 (10-fold cross-validation) | 255 | AUC = 0.79 for radiomics + clinical model vs. AUC = 0.67 for clinical model. |
| Chakraborty et al. [22] | CT | In-house software in MATLAB | BD-IPMN risk | manual (1) | 103 (10-fold cross-validation) | 150 | AUC = 0.77 for radiomics model and AUC = 0.81 for combined radiomics and clinical model. |
| Cheng et al. [29] | CT and MRI | ITK-SNAP software and Artificial Intelligence Kit software |
predicting the malignant potential of intraductal papillary mucinous neoplasms (IPMNs) |
manual (2) | 60 | 1037 | MRI radiomics models achieved improved AUCs (0.879 with LR and 0.940 with SVM, respectively), than that of CT radiomics models (0.811 with LR and 0.864 with SVM, respectively). All radiomics models provided better predictive performance than the clinical and imaging model (AUC = 0.764). |
| Cui et al. [26] |
MRI | MITK software | Low vs. high-grade in BD-IPMNs | manual (2) | 103 + 48/51 (validation1/validation2) | 328 | Radiomics model: AUC = 0.836 (training); AUC = 0.811 (validation1); AUC = 0.822 (validation 2). Radiomics + clinical model: AUC = 0.903 (training); AUC = 0.884 (validation1); AUC = 0.876 (validation 2). |
| D‘Onofrio et al. [30] |
MRI | MevisLasb and MATLAB | Identification and classification of IPMNs |
manual (1) | 91 | <20 | Entropy of the ADC map was found to correlate with tumor dysplasia (p = 0.034, AUC = 0.729) |
| Hanania et al. [17] | CT | IBEX | High-grade vs. low-grade IPMNs | Manual (2) | 53 | 360 | Best univariate AUC = 0.82 |
| Harrington et al. [23] | CT | In-house software in MATLAB | IPMN risk | manual (1) | 33 | <20 | AUC = 0.74 (cyst fluid inflammatory markers model) vs. AUC = 0.83 (radiomics model) vs. AUC = 0.91 (tumor-associated neutrophils model) |
| Huang et al. (2021) [32] | CT | Pyradiomics | Invasiveness of SPN | Manual (2) | 85 | 1316 | Best AUC = 0.914 on 3D-arterial model (compared vs. 2D and venous) |
| Polk et al. [28] | CT | Healthmyne | Malignancy of IPMNs | semi-automatic (1, Healthmyne software) | 51 (5-fold cross-validation) | 39 | AUC = 0.87 (arterial model) vs. AUC = 0.83 (venous model) vs. AUC = 0.90 (combined) |
| Tobaly et al. [18] |
CT | Pyradiomics | Differentiating IPMN grades | Manual (1) | 296 + 112 | 107 | AUC = 0.84 in training set and AUC = 0.71 in validation |
| Wei et al. [20] |
CT | unknown | Computer-aided diagnosis of SCN | Manual (2) | 200 + 60 | 385 | AUC = 0.767 in training and AUC = 0.837 in validation |
| Xie et al. [21] |
CT | In-house algorithm in MATLAB | Differentiating MCN vs. MaSCA | Manual (1) | 57 | 1942 | AUC = 0.989 (radiomics model) vs. AUC = 0.775 (radiological model) vs. AUC = 0.994 (combined model) on bootstrapping |
| Xie et al. [27] |
CT | Pyradiomics | MCN vs. ASCN | semi-manually (1, 3D Slicer) | 216 (10-fold cross-validation) | 764 | Average AUC = 0.784 (radiomics model) vs. AUC = 0.734 (clinical model) |
| Yang et al. [37] |
CT | LifeX | Differentiating SCA vs. MCA | manual (2) | 78 (4:1) | unknown | Slice thickness = 2 mm: AUC = 0.77 in training and AUC = 0.66 in validation; Slice thickness = 5 mm: AUC = 0.72 in training and AUC = 0.75 in validation |
Abbreviations used in this table: Atypical Serous Cystadenomas (ASCN), Branch-Ductal Intraductal Papillary Mucinous Neoplasm (BD-IPMN), Intraductal Papillary Mucinous Neoplasm (IPMN), Macrocystic Serous Cystadenoma (MaSCA), Mucinous Cystadenomas (MCA), Mucinous Cystic Neoplasm (MCN), Neuroendocrine Tumor (NET), Pancreatic Neuroendocrine Neoplasm (PanNEN), Serous Cystadenomas (SCA), Serous Cystic Neoplasms (SCN), Solid Pseudopapillary Neoplasm (SPN).