Table 1 Review of the papers that applied deep learning for brain vessel segmentation.
References | Modality | Data source | No. of subj. | ML model | Input | DSC | Add’l perf. metrics |
|---|---|---|---|---|---|---|---|
Livne et al.48 | TOF MRA | Private (PEGASUS) | 66 | U-Net | 2D patches | 0.88 | 95HD = \(\sim\) 47 voxels an AVD = \(\sim\) 0.4 voxels |
Phellan et al.49 | TOF MRA | Private | 5 | 2D CNN | 2D patches | Between 0.764 0.786 | – |
Hilbert et al.50 | TOF MRA | Private (PEGASUS+ 7UP+ 1000Plus) | 264 | 3D CNN (BRAVE-NET) | 3D patches | 0.931 | 95HD = 29.153, AVD = 0.165 |
Patel et al.51 | DSA | Private | 100 | 3D CNN (DeepMedic) 3D U-Net | 3D patches | 0.94 ± 0.02 0.92 ± 0.02 respectively | CAL = 0.84±0.07 0.79 ± 0.06 respectively |
Tetteh et al.52 | TOF MRA \(\mu\)CTA | Publicly available | Synthetic data | NN with 2D orthogonal cross-hair filters (DeepVesselNet) | 3D volumes | 0.79 | Prec = 0.77 Recall = 0.82 |
Garcia et al.53 | 3DRA | Private | 5 | 3DUNet-based architectures | 3D patches | 0.80 ± 0.06 | Prec = 0.75 ± 0.10 and Recall = 0.90 ± 0.07 |
Vos et al.54 | TOF-MRA | Private | 69 | 2D and 3D U-Net | 2D and 3D patches | 0.74 ± 0.17 0.72 ± 0.15 respectively | MHD = 47.6 ± 40.4 5 81.3 ± 57.0 respectively |
Chatterjee et al.55 | TOF MRI | Not explicity stated | 11 | U-Net MSS | 3D patches | 0.79 ± 0.091 | IOU = 65.89 ± 1.25 |
Zhang and Chen56 | TOF-MRA | Not explicity stated | 42 | DD-Net | 3D patches | 0.67 | Sensitivity = 67.86 IOU = 33.66 |
B.Zhang et al.57 | TOF-MRA | Publicly available (MIDAS) | 109 | DD-CNN | 3D patches | 0.93 | PPV = 96.4730, Sensitivity = 90.1443, Acc = 99.9463 |
Lee et al.58 | MRA | Private (SNUBH) | 26 | 2D U-Net with LSTM (Spider U-Net) | ‘2D images | 0.793 | IOU = 74.3 |
Quintana et al.59 | TOF-MRA | Private (UNAM) | 4 | Dual U-Net-Based cGAN | 2D images | 0.872 | Prec = 0.895 |
