Coronary artery calcification and cardiovascular outcome as assessed by intravascular OCT and artificial intelligence

Abstract

Coronary artery calcification (CAC) is a marker of atherosclerosis and is thought to be associated with worse clinical outcomes. However, evidence from large-scale high-resolution imaging data is lacking. We proposed a novel deep learning method that can automatically identify and quantify CAC in massive intravascular OCT data trained using efficiently generated sparse labels. 1,106,291 OCT images from 1,048 patients were collected and utilized to train and evaluate the method. The Dice similarity coefficient for CAC segmentation and the accuracy for CAC classification are 0.693 and 0.932, respectively, close to human-level performance. Applying the method to 1259 ST-segment elevated myocardial infarction patients imaged with OCT, we found that patients with a greater extent and more severe calcification in the culprit vessels were significantly more likely to have major adverse cardiovascular and cerebrovascular events (MACCE) (p < 0.05), while the CAC in non-culprit vessels did not differ significantly between MACCE and non-MACCE groups.

Fig. 1.

Overview of AI-based automated CAC assessment method. All possible CACs were segmented from OCT pullbacks using the proposed Ca-Net trained with sparsely annotated data, which were then screened and classified by a 3D CNN to reduce false positives. Based on the segmented CACs, clinically relevant metrics were calculated for clinical decision making and prognosis of clinical events.

Fig. 2.

Study population and design. A total of 1,106,347 images from 1,048 patients were collected and sparsely annotated at 1 mm intervals to train and evaluate the model. Another retrospective cohort containing 1259 patients with STEMI was built to evaluate the association between CAC features and clinical outcomes.

Fig. 3.

Proposed Ca-Net for coronary artery calcification segmentation in OCT images. The input images for Ca-Net were five adjacent grayscale Cartesian frames each with 512 $\times$ 512 pixels. (TFC: temporal feature compression module; ReLU: rectified linear unit; BN: batch normalization.)

Fig. 4.

Visual comparisons of Ca-Net and other models on six consecutive OCT images with CACs. Our proposed Ca-Net leverages five contiguous OCT grayscale frames with 512 $\times$ 512 pixels to predict CAC in the middle frame.

Fig. 5.

AI-based CAC assessment model performance at internal evaluation. (A) Image-level and (B) pixel-level performance between the AI method and analysts on 100 pullbacks independently annotated by three analysts. (C) An example of combined CAC assessment with the measurement of CAC angle, thickness, length, depth, and calcium score of an entire OCT pullback. The calcium score [3] was defined as 2 points for calcium angle >180 degrees, 1 point for maximum calcium thickness >0.5 mm, and 1 point for calcium length >5 mm. (Acc: accuracy, Pre: precision, Rec: recall, F1s: F1-score, AI: algorithm, A1: analyst 1, A2: analyst 2, A3: analyst 3, and As: consensus of the three analysts, DSC: Dice similarity coefficient).

Fig. 6.

Prognostic values of various AI-computed CAC features for MACCE. Kaplan-Meier cumulative incidence curves of MACCE stratified by the optimal cutoff of (A) total volume of culprit CAC, (B) maximum thickness of culprit CAC, (C) maximum arc of culprit CAC, and (D) minimum depth of culprit CAC. All optimal cutoffs for calcification features to identify patients at risk of MACCE were determined by Youden’s J statistic on time-dependent ROC curve. CAC: coronary artery calcification; ROC: receiver operating characteristic curve.