Table 1.
Literature on deep learning methods for organ volume measurements in ADPKD.
| First Author |
Modality | Year | ADPKD subjects | Segmentation Methodology |
Dice Score | Other Metrics | Organ |
|---|---|---|---|---|---|---|---|
| Sharma [26] | CT | 2017 | 125 | 2D VGG-16 FCN | 0.86 | 7.8% | Kidney |
| Keshwani [24] | CT | 2018 | 203 CT scans ** | Multi-task 3D FCN | 0.95 | 3.86% | Kidney |
| Shin [27] | CT | 2020 | 214 | 3D V-net | 0.961 * | 95% within 3% | Kidney + Liver |
| Onthoni [25] | CT | 2020 | 97 | 2D SSD Inception Network V2 | - | Images: mAP: 94% Subjects: mAP: 82% |
Kidney |
| Hsiao [23] | CT | 2022 | 210 | FPN + EfficientNet | 0.969 | - | Kidney |
| Jagtap [28] | US | 2022 | 22 | 2D U-Net | 0.80 | 4.12% | Kidney |
| Kim [29] | MRI Cor T2 fatsat |
2016 | 60 | SPPM + PSC | 0.88 | MCC: 0.97 | Kidney |
| Kline [30] | MRI Cor T2 |
2017 | 2000 scans ** | 2D U-Net + ResNet-like encoder |
0.97 | 0.68% | Kidney |
| Guangrui [31] | MRI Axial + Cor T1 |
2019 | 305 | 3D VB-Net *** | RK-0.958 LK-0.965 |
- | Kidney |
| Van Gastel [32] | MRI Cor T2 fatsat |
2019 | 145 | 2D U-Net | - | LK: 0.96 RK: 0.95 TKV: 0.96 Liver: 0.95 |
Kidney + Liver |
| Kline [33] | MRI Cor T2 +/−fatsat |
2020 | 60 | 2D U-Net + ResNet-like encoder |
1st Reader: 0.86 2nd Reader: 0.84 |
1st Reader: 3.9% 2nd Reader: 8% |
Kidney cysts |
| Goel [22] | MRI Axial T2 |
2022 | 173 | 2D U-Net + EfficientNet encoder |
External: 0.98 Prospective: 0.97 |
External: 2.6% Prospective: 3.6% |
Kidney |
| Raj [34] | MRI Cor T1 |
2022 | 100 | 2D Attention U-Net | 0.922 | MSSD: 0.922 and 1.09 mm | Kidney |
| Taylor [35] | MRI | 2022 | 227 Scans | 3D U-Net | 0.96 each kidney |
LK:1.8% RK:1.79% |
Kidney |
FPN = Feature Pyramid Network; FCN = Fully Convolutional Network; SSD = Single Shot Detector; MSSD = Mean Symmetric Surface Distance; RK= right kidney; LK = left kidney; Cor = Coronal; MAPE= Mean absolute percentage error; mAP= mean Average Precision; MCC = mean correlation coefficient. * DSC corresponds to combination of TKV and liver volume. ** number of subjects is unknown. *** customization of V-Net.