Intelligent Detection of Carotid Plaque and Its Stability Based on Deep Learning Dynamic Ultrasound Scanning
- Conditions
- Carotid Atherosclerosis
- Interventions
- Diagnostic Test: Deep learning training cohortDiagnostic Test: Deep learning validation cohort
- Registration Number
- NCT05230576
- Lead Sponsor
- Jia Liu
- Brief Summary
This study intends to build a model through deep learning that can automatically and accurately detect plaques, calculate the lumen stenosis rate and evaluate the stability of plaques based on the carotid transverse axis dynamic ultrasound images and contrast-enhanced ultrasound images, so as to comprehensively evaluate the possibility of carotid plaques. cardiovascular risk. The successful development of this study will automatically simulate and reproduce the whole process of carotid plaque assessment by clinical sonographers. Solve the problem of ultrasonic inspection equipment and experience dependence. It is expected to carry out large-scale population intelligent screening, providing new ideas for early prevention and treatment. Especially in medically underdeveloped remote areas and the lack of experienced sonographers, it has great practical value in clinical health care and can bring greater social and economic benefits.
- Detailed Description
Background:
Carotid plaque is harmful to human health. According to estimates by the World Health Organization, 6.7 million cerebrovascular accidents and strokes occur each year, mainly related to the formation of carotid atherosclerotic plaques. On the one hand, carotid artery plaque can cause carotid artery stenosis or even occlusion, causing cerebral ischemia. Early detection and accurate assessment of carotid plaques are helpful for clinicians to take effective intervention measures, which can significantly reduce the disability rate and fatality rate of stroke.
Carotid CTA and MRA can provide relatively high-resolution and high-quality plaque images, but have cost and scanning limitations that limit their application in daily clinical practice. Ultrasonography has the advantages of non-invasiveness, convenience, low cost, and good repeatability. It is the preferred imaging method for plaque detection, stenosis and plaque stability. Contrast-enhanced ultrasonography (CEUS) can sensitively demonstrate intra-plaque microcirculation perfusion by injecting microbubble contrast agents, and is consistent with histopathological findings, and has been increasingly used clinically to evaluate plaque stability.
However, on the one hand, the limitation of ultrasound examination is that it needs to rely on the level of instruments and operators to improve the accuracy. On the other hand, with the growth of the population base and the aging of society, the traditional medical model has been unable to meet the annual increase in the number of patients. examination needs of patients. Therefore, it is of great significance to develop an integrated AI application platform that can automatically and accurately detect plaque based on ultrasound image data, and evaluate lumen stenosis and plaque stability.
Purpose:
This study intends to build a model based on deep learning to automatically and accurately detect plaque based on the carotid transverse axis dynamic ultrasound image, calculate the lumen stenosis rate, and perform stability assessment, so as to comprehensively evaluate the possible cardiovascular effects of carotid plaque. risk. It will realize the automatic simulation and reproduction of the whole process of assessment of cervical plaque by clinical ultrasound experts.
Study design:
Two-thirds of the enrolled patients and their corresponding carotid artery dynamic scan images and expert diagnosis results were randomly selected as the deep learning training cohort. The carotid artery dynamic scan images and expert diagnosis results of the remaining 1/3 patients were used as a validation cohort to evaluate the overall diagnostic accuracy of the deep learning model
Statistical Analysis:
The sensitivity, specificity, positive predictive value, and negative predictive value of deep learning for detecting plaque, estimating luminal stenosis rate, or predicting plaque stability were calculated by the area under the receiver operating characteristic (ROC) curve (AUROC) to evaluate. Statistical analysis was performed using SPSS 22.0 software.
Quality Control:
Develop standardized and standard carotid ultrasound examination methods and operating procedures, and develop unified image acquisition and storage standards. All operators are rigorously trained in carotid ultrasonography. Two operators with more than 5 years of experience in ultrasound operation were hired as quality control personnel to review all images and exclude unqualified images.
Ultrasound is safe and radiation-free. During the examination, the doctor and the patient were always in a state of communication, and the patient felt less nervous and fearful, with good tolerance and high compliance.
Ethics of the study:
This research will follow the ethical guidelines of the Declaration of Helsinki of the World Medical Congress and the relevant norms and regulations of clinical research. The study will begin after the approval of the ethics committee. Before the start of the study, the investigator should inform the subjects of all relevant contents of the clinical study in easy-to-understand language, and inform the patients that they have the right to withdraw from the study at any time. The study was started only after the patients signed the informed consent voluntarily.
Recruitment & Eligibility
- Status
- COMPLETED
- Sex
- All
- Target Recruitment
- 2000
- (1) Age≥18 years old, gender is not limited. (2) Patients who voluntarily participated in this study signed the informed consent.
- (1) Severe cerebrovascular disease, uncooperative patients, and those who cannot tolerate examination. (2) Wound dressings after neck surgery affects carotid artery ultrasonography. (3) The neck is short and thick, and the probe cannot be put down vertically.
Study & Design
- Study Type
- OBSERVATIONAL
- Study Design
- Not specified
- Arm && Interventions
Group Intervention Description Deep learning training cohort Deep learning training cohort 2/3 of the enrolled patients and their corresponding carotid artery dynamic scan images and expert diagnosis results were randomly selected as the training cohort for deep learning. Deep learning validation cohort Deep learning validation cohort The carotid artery dynamic scan images and expert diagnosis results of the remaining 1/3 patients were used as a validation cohort to evaluate the overall diagnostic accuracy of the deep learning model.
- Primary Outcome Measures
Name Time Method AI assists junior radiologists to read images, and primary physicians read images independently through study completion, an average of 2 years Taking the reading results of senior sonographers as the gold standard, the sensitivity, specificity, accuracy, positive predictive value, and negative predictive value of AI-assisted reading and independent reading by junior physicians for carotid plaque-assisted diagnosis were tested. AUC is evaluated.
AI predicts plaque stability. through study completion, an average of 2 years Taking the reading results of senior sonographers as the gold standard, AI predicts the sensitivity, specificity, accuracy, positive predictive value and negative predictive value of plaque stability. It was evaluated by the area under the receiver operating characteristic (ROC) curve (AUC).
Assessing the performance of AI model through study completion, an average of 2 years Taking the reading results of senior sonographers as the gold standard, the sensitivity, specificity, accuracy, positive predictive value, and negative predictive value of AI independent reading. It was evaluated by the area under the receiver operating characteristic (ROC) curve (AUC).
AI estimates the lumen stenosis rate through study completion, an average of 2 years Taking the reading results of senior sonographers as the gold standard, AI can estimate the sensitivity, specificity, accuracy, positive predictive value and negative predictive value of lumen stenosis rate. It was evaluated by the area under the receiver operating characteristic (ROC) curve (AUC).
Plaque detection by AI model on videos acquired by different types of equipment. through study completion, an average of 2 years Taking the reading results of senior sonographers as the gold standard, AI detects plaque sensitivity, specificity, accuracy, positive predictive value, and negative predictive value on different ultrasound equipment. Assessed by the area under the receiver operating characteristic (ROC) curve (AUC).
- Secondary Outcome Measures
Name Time Method
Trial Locations
- Locations (1)
The Third Affiliated Hospital of Sun Yat-sen University
🇨🇳Guangzhou, Guangdong, China