Deep Learning-Based Detection of Carotid Plaques Informs Cardiovascular Risk Prediction and Reveals Genetic Drivers of Atherosclerosis

Abstract

Atherosclerotic cardiovascular disease, the leading cause of global mortality, is driven by lipid accumulation and plaque formation within arterial walls. Carotid plaques, detectable via ultrasound, are a well-established marker of subclinical atherosclerosis. In this study, we trained a deep learning model to detect plaques in 177,757 carotid ultrasound images from 19,499 UK Biobank (UKB) participants (aged 47-83 years) to assess the prevalence, risk factors, prognostic significance, and genetic architecture of carotid atherosclerosis in a large population-based cohort. The model demonstrated high performance metrics with accuracy, sensitivity, specificity, and positive predictive value of 89.3%, 89.5%, 89.2%, and 82.9%, respectively, identifying carotid plaques in 45% of the population. Plaque presence and count were significantly associated with future cardiovascular events over a median follow-up period of up to 7 years, leading to improved risk reclassification beyond established clinical prediction models. A genome-wide association study (GWAS) meta-analysis of carotid plaques (29,790 cases, 36,847 controls) uncovered two novel genomic loci (p < 5×10^-8) with downstream analyses implicating lipoprotein(a) and interleukin-6 signaling, both targets of investigational drugs in advanced clinical development. Observational and Mendelian randomization analyses showed associations between smoking, low-density-lipoprotein (LDL) cholesterol, and high blood pressure and the odds of carotid plaque presence. Our study underscores the potential of carotid plaque assessment for improving cardiovascular risk prediction, provides novel insights into the genetic basis of subclinical atherosclerosis, and offers a valuable resource for advancing atherosclerosis research at the population scale.

Keywords: Mendelian Randomization; atherosclerosis; cardiovascular disease; carotid artery; genetics; machine learning; vascular ultrasound.

Figure 1.. Summary of the study design.

ASCVD – Atherosclerotic Cardiovascular Disease, CVD – Cardiovascular disease, PCE – Pooled Cohort Equatios, TP– true positive, FN – False Negative, FP – False Positive, TN – True Negative, GWAS – Genome-Wide Association Study

Figure 2.. Development and performance of the plaque detection model.

A. General representation of the YoloV8 architecture, C – concatenation, C2F – cross-stage partial bottleneck with two convolutions, U – up-sampling, Conv – convolutional module, P1-P5 represent future maps; Bbox – bounding box prediction branch, Cls – classification branch, BCE – Binary Cross Entropy loss, DFL – Distribution Focal Loss, CIoU – Complete Intersection over Union, nc – number of classes, reg_max – maximum value for the bounding box regression. Adopted from Terven & Cordova-Esparza and https://github.com/ultralytics/ultralytics/issues/189. B. Confusion matrix for model’s classification performance for plaque presence at each image. The confusion matrix is based on comparing the presence of plaque in the annotations with the model’s predictions. Specifically, if there is a plaque annotation for an image and the prediction contains a bounding box, then the prediction is annotated as true positive. C. Examples of model predictions as compared to manual annotations for three UK Biobank participants in the test set. Each image depicts the longitudinal view of the common carotid artery extending toward the bifurcation area (the left part of the images).

Figure 3.. Prevalence and predictors of carotid plaque in the UK Biobank.

A. Distribution of plaques in the left and right carotid arteries. B. Percentage of plaque presence across age and sex groups. The error bars represent 95% confidence intervals. С. Forest plot of the associations of demographic and vascular risk factors with the odds of carotid plaque presence, as derived by a multivariable logistic regression model that includes all variables in the figure. The results are presented as odds ratios (OR) and 95% confidence intervals (CI). SBP – systolic blood pressure; CVD – cardiovascular disease; LDL – low-density lipoprotein cholesterol; SD – standard deviation. * History of hypertension is defined by the use of antihypertensive medication

Figure 4.. Survival curves of cumulative major adverse cardiovascular event (MACE) rates by total count of carotid plaques predicted by the model.

The presented hazard ratios (HRs) were estimated using Cox regression, adjusted for: sex, age, systolic blood pressure, use of statins, history of antihypertensive therapy, current smoking, history of diabetes, HDL cholesterol and total cholesterol levels.

Figure 5.. Manhattan plot of the GWAS meta-analysis for carotid plaque presence (29,790 cases; 36,847 controls).

Loci highlighted in red point to novel significant associations for carotid plaque whereas loci highlighted in green represent validation of previously described associations.

Figure 6.. Forest plot for Mendelian randomization results.

The results are presented as odds ratios (OR) and 95% confidence intervals (CI) derived from random-effects inverse-variance weighted Mendelian randomization analyses. Two-fold increments in prevalence for binary exposures (type 2 diabetes and smoking initiation) were derived by multiplying the IVW betas and corresponding confidence intervals by 0.693, as described by Burgess and Labrecque. HbA1c – Glycated hemoglobin; BMI – Body Mass Index; LDL – Low-Density Lipoprotein Cholesterol; HDL – High-Density Lipoprotein Cholesterol; SBP – Systolic Blood Pressure; DBP – Diastolic Blood Pressure; IL-6 – Interleukin-6.