Table 5.
A summary of reviewed deep learning methods for ultrasound image segmentation.
| Application | Selected works | Method | Structure | Imaging modality |
|---|---|---|---|---|
| Combined with deformable models | ||||
| Carneiro et al. (138, 139) | DBN with two-step approach: localization and fine segmentation | LV | 2D A2C, A4C | |
| Nascimento and Carneiro (140) | deep belief networks (DBN) and sparse manifold learning for the localization step | LV | 2D A2C, A4C | |
| Nascimento and Carneiro (141, 142) | DBN and sparse manifold learning for one-step segmentation | LV | 2D A2C, A4C | |
| Veni et al. (143) | FCN (U-net) followed by level-set based deformable model | LV | 2D A4C | |
| Utilizing temporal coherence | ||||
| 2D LV | Carneiro and Nascimento (144, 145) | DBN and particle filtering for dynamic modeling | LV | 2D A2C, A4C |
| Jafari et al. (146) | U-net and LSTM with additional optical flow input | LV | 2D A4C | |
| Utilizing unlabeled data | ||||
| Carneiro and Nascimento (147, 148) | DBN on-line retrain using external classifier as additional supervision | LV | 2D A2C, A4C | |
| Smistad et al. (149) | U-Net trained using labels generated by a Kalman filter based method | LV and LA | 2D A2C, A4C | |
| Yu et al. (150) | Dynamic CNN fine-tuning with mitral valve tracking to separate LV from LA | Fetal LV | 2D | |
| Jafari et al. (151) | U-net with TL-net (152) based shape constraint on unannotated frames | LV | 2D A4C | |
| Utilizing data from multiple domains | ||||
| Chen et al. (153) | FCN trained using annotated data of multiple anatomical structures | Fetal head and LV | 2D head, A2-5C | |
| Others | ||||
| Smistad et al. (154) | Real time CNN view-classification and segmentation | LV | 2D A2C, A4C | |
| Leclerc et al. (155) | U-net trained on a large heterogeneous dataset | LV, Myo | 2D A4C | |
| Jafari et al. (156) | Real-time mobile software, lightweight U-Net, multitask and adversarial training | LV | 2D A2C, A4C | |
| Dong et al. (157) | CNN for 2D coarse segmentation refined by 3D snake model | LV | 3D (CETUS) | |
| 3D LV | Oktay et al. (59) | U-net with TL-net based shape constraint | LV | 3D (CETUS) |
| Dong et al. (158) | Atlas-based segmentation using DL registration and adversarial training | LV | 3D | |
| Ghesu et al. (159) | Marginal space learning and adaptive sparse neural network | Aortic valves | 3D | |
| Others | Degel et al. (160) | V-net with TL-net based shape constraint and GAN-based domain adaptation | LA | 3D |
| Zhang et al. (42) | CNN for view-classification, segmentation and disease detection | Multi-chamber | 2D PLAX, PSAX, A2-4C | |
A[X]C is short for Apical [X]-chamber view. PLAX/PSAX, parasternal long-axis/short-axis; CETUS, using the dataset from Challenge on Endocardial Three-dimensional Ultrasound Segmentation.