Key result
Deep learning on carotid ultrasounds predicts MI risk with 98% accuracy.
Why the study?
Carotid Intima-Media Thickness (CIMT) can support estimation of cardiovascular and heart failure risk, prompting evaluation of CIMT and Framingham risk score boundaries using machine learning and deep learning.
Can machine learning and deep learning techniques using CIMT and FRS accurately predict the risk of myocardial infarction?
Cross-Sectional (n=110)
Can machine learning and deep learning techniques using CIMT and FRS accurately predict the risk of myocardial infarction?
Deep learning models using carotid intima-media thickness from ultrasound images can predict myocardial infarction risk with high accuracy.
DL models on carotid ultrasound may refine MI risk stratification in practice; extends FRS-CIMT correlations but requires larger validation.
Myocardial infarction (MI) is a life threatening condition that causes death in developing nations due to blockage in the blood vessel of the coronary artery that supplies blood to the heart. Any blockage in blood vessel of carotid artery located in the neck region can predict the risk of heart failure. Carotid Intima-Media Thickness (CIMT) measured from carotid artery supports in the estimation of heart failure. In this study, cardiovascular risk is considered based on the CIMT thickness of carotid artery blood vessels. In this paper, CIMT and Framingham risk score (FRS) boundary have been determined for both normal and cardiovascular disease (CVD) subjects, which aids in the prediction of heart failure. For 55 subjects with normal condition and 55 subjects with CVD disease, CIMT values were measured by utilizing an effective ultrasound examination system. Biochemical parameters were also measured for all the 110 subjects to predict the FRS score. Student t-test and Spearmans correlation performed showed significant results with p-value less than 0.01 for CIMT, biochemical parameters and FRS score. Receiver operating characteristic (ROC) has been plotted for measured CIMT value and FRS value indicates with an accuracy of 71%. The performance was also determined by comparing various classification techniques using machine learning and deep learning. Results observed through machine learning show that random forest, multilayer perceptron and K-nearest neighbor classifiers used in classification techniques give more superior accuracy of 79% and sensitivity of 78%. To improvise the investigation, deep learning technique using carotid artery ultrasound image of 1909 dataset has been used. Deep learning CNN architecture using VGG19 implemented gave an accuracy of 98% with sensitivity 98% and specificity 99%. It was also observed from the 20% data used for validation that 199 subjects are without risk of MI and 178 subjects were predicted with a risk of MI in future. Further relative risk analysis performed with FRS and CIMT showed that a person with low risk of FRS has 26% chances of getting abnormal CIMT and MI risk in future.
No takes yet. Share an insight, caveat, or question.
Prabha et al. (2022) conducted a cross-sectional in Myocardial infarction risk (n=110). Deep learning CNN architecture (VGG19) vs. Machine learning classifiers (random forest, MLP, KNN) was evaluated on Classification accuracy. Deep learning using VGG19 architecture on carotid artery ultrasound images predicted myocardial infarction risk with 98% accuracy, 98% sensitivity, and 99% specificity.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: