Assessment of Stroke Risk Using MRI-VPD with Automatic Segmentation of Carotid Plaques and Classification of Plaque Properties Based on Deep Learning

Yi Wang,Chi Cai,Ye-Ming Du,Yuan-Zhe Li,Shu-Ting Li,Jing Huang,Hong Wu
DOI: https://doi.org/10.1016/j.jrras.2023.100630
IF: 1.77
2023-01-01
Journal of Radiation Research and Applied Sciences
Abstract:Purpose: The primary goal of this study is to establish a technique for categorizing MRI scans into two distinct groups, namely, vulnerable and stable carotid plaques, for the purpose of evaluating risk of stroke levels based on imaging of carotid plaque.Method: The study data was obtained from two hospitals and included cases of both stable and vulnerable conditions. The cohort comprised of 87 patients with established risk factors for atherosclerosis. A total of 3741 MRI scans were utilized, with each category evenly represented and randomly split into training and testing sets at a ratio of 70:30, respectively. Our methodology utilized the U-Net model for automatic extraction of the region of interest (ROI) from the MRI scans and subsequent plaque image segmentation, followed by the imple-mentation of pre-trained ResNet-50 models for classification.Results: Our proposed framework involved the integration of two deep learning models, namely U-net and ResNet-50, with hyperparameters fine-tuned and adjusted based on our classification problem. We attained the highest accuracy of 94.11% using our proposed framework, thus highlighting the efficacy of our approach. Conclusion: Our study indicates that pre-trained deep learning models have the potential to classify MRI scans based on carotid plaque imaging with fine-tuning. The proposed framework can help reduce erroneous diagnoses caused by suboptimal image quality or individual expertise. Furthermore, our framework can assist radiologists in making better decisions for patient care. In conclusion, this study provides a useful tool for the assessment of stroke risk levels, which can ultimately improve patient outcomes.
What problem does this paper attempt to address?