Author Affiliations
1Department of Electronic and Communication Engineering, North China Electric Power University, Baoding 071003, Hebei , China2Hebei Key Laboratory of Power Internet of Things Technology, North China Electric Power University, Baoding 071003, Hebei , Chinashow less
Fig. 1. Original IVOCT image and comparison of segmentation results of five methods
[19]. (a) Original image; (b) manual segmentation by specialists; (c) segmentation by prototype U-Net; (d) segmentation by prototype U-Net with proposed loss function; (e) segmentation by deep Residual U-Net and ResNet101; (f) segmentation by deep Residual U-Net and ResNet101 with proposed loss function
Fig. 2. Illustration of SegNet architecture
[20] Fig. 3. Segmentation results of SegNet
[22]. (a) Original image; (b) ground truth; (c) initial segmentation by SegNet; (d) output after CRF processing
Fig. 4. Segmentation results of IVOCT images by DeepCap
[17] Fig. 5. Segmentation results of lumen contour in an IVOCT image with metal stent
[24].
(a) Original image; (b) ground truth; (c) U-Net output; (d) N-Net output
Fig. 6. Segmentation results of IVOCT lumen contour
[25].
(a) Original images; (b) ground truth; (c) U-Net output; (d) FCN output; (e) SegNet output; (f) RSM-Network output
Fig. 7. Comparison of segmentation results of vessel border from IVOCT images by different methods
[26]. (a) Vessel bifurcation; (b) white thrombus; (c) red thrombus; (d) complex thrombus
Fig. 8. Polar views of IVOCT images after tissue characterization where the patches with red represent plaque tissue and those with green are normal vessel wall
[48]. (a) Lipid plaque; (b) calcified and fibrous plaque; (c) mixed plaque; (d) calcified plaque
Fig. 9. Classification results for each coronary artery tissue type in polar views of IVOCT images
[50]. (a) Intima; (b) media; (c) fibrosis
Fig. 10. Detection result of metallic stents in IVOCT images
[61]. (a) YOLOv3; (b) R-FCN
Fig. 11. Results of stent detection in IVOCT images
[65]. (a) No stent; (b) metal stent; (c) BVS
Fig. 12. Flowchart of ML-based method for automatic identification of vascular bifurcations
[67] Fig. 13. CNN architecture used to classify bifurcation regions in IVOCT images
[69] Method | Advantage | Disadvantage |
---|
U-Net | Having simple structure and easy to operate; Capable of incorporating features of different scales | Image details might be lost as the network deepens | SegNet | Having fewer parameters and capable of preserving high-frequency details | Incapable of segmenting objects with complex categories | CapsNet | Requiring less training data; capable of processing with high speed; capable of preserving a large amount of image details | Having high computational burden; Incapable of accurately distinguishing objects close to each other; having outstanding performance only on small data sets | N-Net | Capable of fusing multiscale feature maps | Segmentation accuracy needs to be further improved | RSM-Network | Having strong ability of learning detail features; capable of segmenting boundaries with high robustness and accuracy | Having bad performance in segmenting hierarchical features | BPN | Capable of effectively extracting the semantic features of vessels and preserving the high-resolution image information of vascular boundaries | Requiring complex training process |
|
Table 1. Comparison of six DL-based methods for IVOCT image segmentation