Intravascular Ultrasound Image Segmentation Fusing Transformer Branch and Topology Enforcement

Wenyue Hao; Huaiyu Cai; Tingtao Zuo; Zhongwei Jia; Yi Wang; Xiaodong Chen

doi:10.3788/LOP231918

Laser & Optoelectronics Progress, Volume. 61, Issue 12, 1237008(2024)

Intravascular Ultrasound Image Segmentation Fusing Transformer Branch and Topology Enforcement

Wenyue Hao¹, Huaiyu Cai^1、*, Tingtao Zuo², Zhongwei Jia³, Yi Wang¹, and Xiaodong Chen¹

¹Ministry of Education Key Laboratory of Optoelectronic Information Technology, School of Precision Instrument and Optoelectronic Engineering, Tianjin University, Tianjin 300072, China

²Lepu Medical Technology (Beijing) Co., Ltd., Beijing 102200, China

³Southwestern Lu Hospital, Liaocheng 252325, Shandong, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(8)

Fig. 1. Schematic diagram of the network structure. (a) Backbone network structure; (b) overall network structure

Download full size

Fig. 2. Self attention calculation methods. (a) Axial self attention calculation^[26]; (b) gated axial self attention calculation^[27]

Download full size

Fig. 3. Compared with the visualization results of other models, where the yellow line on the outside represents the segment of the media and the green line on the inside represents the segment of the lumen

Download full size

Table 1. Comparison results with other models

View table

Table 1. Comparison results with other models

Model	C_Dice（↑）		$M_{J M}$ （↑）		$D_{H D}$ （↓）		$P_{P A D}$ （↓）		$R_{T E R}$ （↓）
Model	Lum	Med	Lum	Med	Lum	Med	Lum	Med	All
UNet	0.929	0.944	0.871	0.896	0.250	0.391	0.088	0.090	10.40%
Deeplabv3+	0.929	0.949	0.871	0.905	0.199	0.186	0.098	0.08	5.96%
TransUNet	0.932	0.936	0.875	0.884	0.203	0.324	0.088	0.104	8.26%
Swin-UNet	0.937	0.951	0.884	0.908	0.300	0.437	0.081	0.064	1.83%
Swin-UNet^*	0.921	0.887	0.858	0.805	0.997	2.259	0.123	0.211	12.80%
Proposed algorithm	0.939	0.953	0.889	0.912	0.102	0.103	0.066	0.072	0%

Table 2. Results of ablation experiment

View table

Table 2. Results of ablation experiment

Model	C_Dice（↑）		$M_{J M}$ （↑）		$D_{H D}$ （↓）		$P_{P A D}$ （↓）		$R_{T E R}$ （↓）
Model	Lum	Med	Lum	Med	Lum	Med	Lum	Med	All
CNN	0.930	0.944	0.873	0.897	0.241	0.367	0.095	0.086	8.7%
Transformer	0.928	0.916	0.869	0.851	0.209	0.354	0.108	0.156	9.4%
CNN+Transformer	0.936	0.953	0.882	0.912	0.108	0.138	0.066	0.078	2.7%
Proposed algorithm	0.939	0.953	0.889	0.912	0.102	0.103	0.066	0.072	0

Table 3. Validity verification of enhanced mixed-FFN in backbone network
View table
Table 3. Validity verification of enhanced mixed-FFN in backbone network
Model C_Dice（↑） $M_{J M}$ （↑） $D_{H D}$ （↓） $P_{P A D}$ （↓）
Lum Med Lum Med Lum Med Lum Med
Gated Transformer 0.925 0.900 0.861 0.827 0.345 0.560 0.114 0.158
Enhanced Transformer 0.928 0.916 0.869 0.851 0.209 0.354 0.108 0.156

Table 4. Segmentation results of different Transformer layers

View table

Table 4. Segmentation results of different Transformer layers

Transformer layers	C_Dice（↑）		$M_{J M}$ （↑）		$D_{H D}$ （↓）		$P_{P A D}$ （↓）
Transformer layers	Lum	Med	Lum	Med	Lum	Med	Lum	Med
0	0.930	0.944	0.873	0.897	0.241	0.367	0.086	0.095
1	0.932	0.945	0.876	0.897	0.220	0.408	0.088	0.083
2	0.932	0.95	0.877	0.906	0.168	0.220	0.069	0.095
3	0.936	0.953	0.882	0.912	0.108	0.138	0.066	0.078
4	0.933	0.946	0.877	0.900	0.165	0.180	0.067	0.089

Table 5. Comparison of different filtering modes in topologically enforced networks
View table
Table 5. Comparison of different filtering modes in topologically enforced networks
Model C_Dice（↑） $M_{J M}$ （↑） $D_{H D}$ （↓） $P_{P A D}$ （↓）
Lum Med Lum Med Lum Med Lum Med
Gaussian smoothing 0.940 0.949 0.889 0.905 0.107 0.115 0.071 0.075
Bilateral smoothing 0.939 0.953 0.889 0.912 0.102 0.103 0.066 0.072

Tools

Get Citation

Copy Citation Text

Wenyue Hao, Huaiyu Cai, Tingtao Zuo, Zhongwei Jia, Yi Wang, Xiaodong Chen. Intravascular Ultrasound Image Segmentation Fusing Transformer Branch and Topology Enforcement[J]. Laser & Optoelectronics Progress, 2024, 61(12): 1237008

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites