Multi-Sequence MRI Registration of Atherosclerotic Carotid Arteries Based on Cross-Scale Siamese Network
Xiaojie Huang1, Lizhao Mao2, Xiaoyan Wang2
1The Second Affiliated Hospital, School of Medicine, Zhejiang University, Hangzhou, China.
Insights
This study introduces a novel Siamese U-Net method for accurately registering multi-sequence carotid MRI scans, improving diagnosis of carotid atherosclerosis (CAS). The advanced technique enhances registration accuracy, aiding in the evaluation of cardiovascular disease progression.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiovascular Medicine
Background:
- Cardiovascular disease (CVD) poses a significant mortality risk, with carotid atherosclerosis (CAS) being a primary contributor.
- Multisequence carotid MRI offers detailed plaque analysis but faces challenges in accurate registration due to parameter inconsistencies and motion artifacts.
- Evaluating local changes in CAS progression from multi-sequence MRI is hindered by spatial misalignment.
Purpose of the Study:
- To develop a robust cross-scale, multi-modal image registration method for multi-sequence carotid MRI.
- To address the challenges of inconsistent parameters and geometric space mismatch in carotid atherosclerosis imaging.
- To enhance the accuracy of diagnosing and monitoring cardiovascular disease through improved MRI analysis.
Main Methods:
- A Siamese U-Net architecture was employed, utilizing sub-networks with varying input sizes for diverse feature extraction.
- A specialized padding module was designed to enable training on cross-scale features.
- A multi-scale loss function incorporating Gaussian smoothing was implemented to optimize registration performance.
Main Results:
- The proposed method achieved a Dice Similarity Coefficient (DSC) of 84.1% on a carotid dataset with cross-size inputs.
- The Gaussian smoothing loss function improved the average DSC by 5.23%.
- The average distance between landmarks decreased by 6.46%, indicating enhanced registration precision.
Conclusions:
- The developed cross-scale Siamese U-Net method provides precise and reliable registration for multi-sequence carotid MRI.
- The integration of Gaussian smoothing loss significantly boosts registration accuracy for carotid atherosclerosis assessment.
- This approach offers a promising tool for improving the diagnostic accuracy and monitoring of cardiovascular disease progression.
Abstract:
Cardiovascular disease (CVD) is a common disease with high mortality rate, and carotid atherosclerosis (CAS) is one of the leading causes of cardiovascular disease. Multisequence carotid MRI can not only identify carotid atherosclerotic plaque constituents with high sensitivity and specificity, but also obtain different morphological features, which can effectively help doctors improve the accuracy of diagnosis. However, it is difficult to evaluate the accurate evolution of local changes in carotid atherosclerosis in multi-sequence MRI due to the inconsistent parameters of different sequence images and the geometric space mismatch caused by the motion deviation of tissues and organs. To solve these problems, we propose a cross-scale multi-modal image registration method based on the Siamese U-Net. The network uses sub-networks with image inputs of different sizes to extract various features, and a special padding module is designed to make the network available for training on cross-scale features. In addition, to improve the registration performance, a multi-scale loss function under Gaussian smoothing is applied for optimization. For the experiments, we have collected a multi-sequence MRI image dataset from 11 patients with carotid atherosclerosis for a retrospective study. We evaluate our overall architectures by cross-validation on our carotid dataset. The experimental results show that our method can generate precise and reliable results with cross-scale multi-sequence inputs and the registration accuracy can be greatly improved by using the Gaussian smoothing loss function. The DSC of our Siamese structure can reach 84.1% on the carotid data set with cross-size input. With the use of GDSC loss, the average DSC can be improved by 5.23%, while the average distance between fixed landmarks and moving landmarks can be decreased by 6.46%.Our code is made publicly available at: https://github.com/MingHan98/Cross-scale-Siamese-Unet.
