Artificial intelligence for the analysis of intracoronary optical coherence tomography images: a systematic review

Abstract

Intracoronary optical coherence tomography (OCT) is a valuable tool for, among others, periprocedural guidance of percutaneous coronary revascularization and the assessment of stent failure. However, manual OCT image interpretation is challenging and time-consuming, which limits widespread clinical adoption. Automated analysis of OCT frames using artificial intelligence (AI) offers a potential solution. For example, AI can be employed for automated OCT image interpretation, plaque quantification, and clinical event prediction. Many AI models for these purposes have been proposed in recent years. However, these models have not been systematically evaluated in terms of model characteristics, performances, and bias. We performed a systematic review of AI models developed for OCT analysis to evaluate the trends and performances, including a systematic evaluation of potential sources of bias in model development and evaluation.

Keywords: Artificial intelligence; Coronary; Deep learning; Intravascular imaging; Optical coherence tomography; Systematic review.

Graphical Abstract

Systematic review on artificial intelligence (AI) tasks in intracoronary optical coherence tomography (OCT): performance and bias signalling. Left panel: Typical tasks for AI for the automated assessment of intracoronary OCT images include classification, object detection, and segmentation. Classification can be performed framewise or on individual A-lines and is typically used in a binary fashion (i.e. presence or absence of the feature of interest) or with multiple classes (e.g. lipidic plaque, calcified plaque, fibrous plaque, etc.). Object detection refers to the task of localizing a feature of interest, and is typically performed with bounding boxes (dashed lines) or a centroid (blue dot). Semantic segmentation refers to pixel-wise labelling of the feature(s) of interest. Quantification is frequently performed using post-processing steps after semantic segmentation or A-line classification and includes for example measurements of plaque components. Middle panel: Among all identified models, the overall performances (reported as medians with minimum and maximum values) for lumen and stent assessment were excellent, while plaque analysis showed varied results. Right panel: The bias signalling questionnaire, containing several questions related to data management, model development, and validation, revealed potential caveats in model training and evaluation that should be addressed in future studies.

Figure 1

Flowchart of study selection.

Figure 2

Annual publications on machine learning and deep learning.

Figure 3

Examples of intracoronary optical coherence tomography images in polar and Cartesian views. The top row presents three optical coherence tomography frames in polar view, while the reconstructed Cartesian view is displayed in the bottom row. (A) Polar and Cartesian view of the same frame illustrating the y-axis aligning with radius (r) and the x-axis corresponding to the angle (Θ). Corresponding A-lines are denoted by an asterisk. (B) Example of an optical coherence tomography frame with a lipidic plaque with a thick fibrous cap (arrows). The inner boundary of the lipidic plaque is outlined with a continuous line, while the estimated outer boundary is represented by a dashed line. (C) Example of an optical coherence tomography frame with a calcified plaque (white outline).

Figure 4

Distribution of tasks per class. Cumulative number of identified tasks for segmentation, classification, object detection, and regression over time (A–C) and per class (D). Quantification is not an artificial intelligence task but refers to transforming the model output into numerical variables to quantify for example plaque volume or fractional flow reserve.

Figure 5

Results of bias signalling questions.

Figure 6

Examples of artificial intelligence algorithms for different tasks. (A) Multiclass semantic segmentation model proposed by Chu et al. The raw optical coherence tomography image is presented in the left image, the reference standard in the middle image, and the model prediction in the right image. White represents calcium, green represents intima/fibrous, yellow represents lipid, and purple represents guidewire artefact. Used with permission of EuroIntervention, from Artificial intelligence and optical coherence tomography for the automatic characterization of human atherosclerotic plaques, Chu, 17, 1, 2021; permission conveyed through Copyright Clearance Center, Inc. (B) Detection model for the localization of metal stent struts using centroids proposed by Yang et al. Adapted with permission from Yang et al. © Optica Publishing Group. (C) Framewise classification for the presence vs. absence of thin-cap fibroatheroma proposed by Min et al. The right panel represents the gradient-weighted class activation mapping (Grad-CAM) with red colours indicating the high attended regions to predict thin-cap fibroatheroma. Used with permission of EuroIntervention, from Detection of optical coherence tomography defined thin-cap fibroatheroma in the coronary artery using deep learning, Min, 16, 5, 2020; permission conveyed through Copyright Clearance Center, Inc.