Self-Supervised Pre-Training for Intravascular Ultrasound Image Segmentation Method Based on Diffusion Model

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

doi:10.3788/LOP232774

Laser & Optoelectronics Progress, Volume. 61, Issue 18, 1837005(2024)

Self-Supervised Pre-Training for Intravascular Ultrasound Image Segmentation Method Based on Diffusion Model

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

¹Key Laboratory of Optoelectronics Information Technology, Ministry of Education, School of Precision Instrument and Opto-Electronics 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(9)

Fig. 1. Schematic diagram of example of encoder-decoder structure^[26]

Download full size

Fig. 2. Denoising self-supervised pre-training paradigm based on diffusion

Download full size

Fig. 3. Comparison of different initialization methods of 20% data set

Download full size

Fig. 4. Comparison of visualization results of segmentation results (yellow line on the outside represents the segment of the media and green line on the inside represents the segment of the lumen)

Download full size

Table 0. [in Chinese]

View table

Table 0. [in Chinese]

算法1基于扩散的去噪自监督代理任务伪代码

输入：

$x_{t}$ ：输入图像

输出：

$ϵ_{θ} (x_{t})$ ：输出图像

算法流程：

1：repeat

2： $x_{0} ~ q (x_{0})$

3： $t ~ U n i f o r m ({1, \dots, T})$

4： $ϵ ~ N (0,1)$

5： take gradient desent step on $\nabla_{θ} | | ϵ_{θ} (\sqrt[]{\bar{α_{t}}} x_{0} + \sqrt[]{1 - \bar{α_{t}}} ϵ) - ϵ | |^{2}$

6：until converged

Table 1. Comparison of results of different initialization methods

View table

Table 1. Comparison of results of different initialization methods

Method	D_Dice（↑）		J_JM（↑）		H_HD（↓）		P_PAD（↓）
Method	Lumen	Media	Lumen	Media	Lumen	Media	Lumen	Media
Scratch（20%）	0.902	0.875	0.831	0.792	0.385	0.450	0.195	0.301
Pred_x（20%）	0.921	0.933	0.857	0.878	0.182	0.252	0.107	0.104
Pred_noise（20%）	0.932	0.942	0.875	0.893	0.169	0.343	0.088	0.082
Scratch（100%）	0.932	0.949	0.877	0.906	0.226	0.268	0.092	0.078

Table 2. Comparison of experimental results using different methods

View table

Table 2. Comparison of experimental results using different methods

Model	D_Dice（↑）		J_JM（↑）		H_HD（↓）		P_PAD（↓）
Model	Lumen	Media	Lumen	Media	Lumen	Media	Lumen	Media
Unet（100%）	0.929	0.944	0.871	0.896	0.250	0.391	0.088	0.090
Deeplabv3+（100%）	0.929	0.949	0.871	0.905	0.199	0.186	0.098	0.080
TransUnet（100%）	0.932	0.936	0.875	0.884	0.203	0.324	0.088	0.104
Swin-Unet（100%）	0.921	0.887	0.858	0.805	0.997	2.259	0.123	0.211
Pred_noise（20%）	0.932	0.942	0.875	0.893	0.169	0.343	0.088	0.082

Table 3. Comparison of results of different noise level numbers

View table

Table 3. Comparison of results of different noise level numbers

T	D_Dice（↑）		J_JM（↑）		H_HD（↓）		P_PAD（↓）
T	Lumen	Media	Lumen	Media	Lumen	Media	Lumen	Media
1	0.919	0.921	0.855	0.860	0.318	0.482	0.121	0.139
100	0.927	0.939	0.866	0.888	0.124	0.181	0.093	0.074
200	0.932	0.942	0.875	0.893	0.169	0.343	0.088	0.082
500	0.926	0.941	0.865	0.891	0.169	0.327	0.088	0.084
1000	0.920	0.936	0.856	0.883	0.367	0.451	0.117	0.094

Table 4. Comparison of results of different loss functions

View table

Table 4. Comparison of results of different loss functions

Loss function	D_Dice（↑）		J_JM（↑）		H_HD（↓）		P_PAD（↓）
Loss function	Lumen	Media	Lumen	Media	Lumen	Media	Lumen	Media
$L_{M S E}$	0.927	0.937	0.866	0.884	0.197	0.306	0.088	0.087
$L_{M S E}$ +0.1 $L_{S S I M}$	0.932	0.942	0.875	0.893	0.169	0.343	0.088	0.082

Tools

Get Citation

Copy Citation Text

Wenyue Hao, Huaiyu Cai, Tingtao Zuo, Zhongwei Jia, Yi Wang, Xiaodong Chen. Self-Supervised Pre-Training for Intravascular Ultrasound Image Segmentation Method Based on Diffusion Model[J]. Laser & Optoelectronics Progress, 2024, 61(18): 1837005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Digital Image Processing

Received: Dec. 28, 2023

Accepted: Feb. 5, 2024

Published Online: Sep. 14, 2024

The Author Email: Huaiyu Cai (hycai@tju.edu.cn)

DOI:10.3788/LOP232774

CSTR:32186.14.LOP232774

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology