Automated classification of coronary atherosclerotic plaque in optical frequency domain imaging based on deep learning
Graphical abstract
Introduction
Optical frequency domain imaging (OFDI) and optical coherence tomography (OCT) are widely used as imaging tools for characterizing coronary atherosclerotic plaques in vivo [[1], [2], [3], [4]]. Previous ex vivo studies have reported that OCT can be used with high sensitivity and specificity [5,6] to differentiate between three types of atherosclerotic plaque components (fibrous, fibrocalcific, and lipid-rich) in autopsy specimens. OFDI/OCT can also detect macrophages, necrotic cores, cholesterol crystals, calcified nodules, and intimal vasculature [7,8]. However, the accuracy of OFDI/OCT for histological diagnosis of atherosclerotic lesions depends on the experience of the observer [9]. The accurate interpretation of OFDI scans can be challenging for less experienced observers.
Deep learning (DL) is a field of artificial intelligence that has received significant attention in recent years [10]. Semantic segmentation in DL is the process of partitioning a digital image into multiple segments. This technique, also known as pixel labeling, is used to extract objects by classifying them on a pixel-by-pixel basis. In this study, we hypothesized that pathological classification from OFDI images can be done by the DL-based segmentation model.
We have presented a new computerized technique for atherosclerotic plaque categorization. The main contribution of our work is improvements in the feature extraction stage, wherein the desired features are determined using auto-correlation in a completely automatic process without the need for time-consuming training.
Section snippets
Study design
We retrospectively collected ex vivo OFDI images of hearts from autopsies. The images were randomly divided into training, validation, and test dataset. The model was trained with the training dataset and tuned with the validation dataset. The developed model was then evaluated using the test dataset that was independent from the training and validation dataset. The study protocol was approved by the Institutional Review Board (2019075). Written informed consent for participation in this study
Data
A total of 105 coronary arteries from 45 consecutive human cadavers were examined to match OFDI and histological images. The median length of coronary artery segments imaged by OFDI pullbacks was 67 mm (range: 12–109 mm). In cases where a coronary stent was involved, the stented segment and its distal and proximal regions (10 mm on each side) were excluded from the analysis. OFDI images of 1693 histological cross-sections were compared. Among these sections, 433 histological segments were
Discussion
Our method for automated semantic segmentation of coronary plaques using OFDI was promising in terms of categorizing the atherosclerotic plaque using histological findings as the gold standard. A DL model on a folded, large, and manually annotated dataset of OFDI images, and their histological diagnosis information was trained and tested. The results showed that the model can provide an accurate histological diagnosis of coronary lesions from a single cross-sectional OFDI image.
Intravascular
Financial support
Dr. Shibutani has received grants from Japan Arteriosclerosis Prevention Fund, and research support from the Uehara Memorial Foundation, Fukuda Foundation for Medical Technology, and the Japan Society for the Promotion of Science Overseas Challenge Program for Young Researchers. This work was financially supported in part by the JSPS KAKENHI Grant Number 21K08065.
Data availability
The source code for the deep learning model is available on GitHub (https://github.com/nine-teen-eight-hundred/c-oct/commit/854c30fd9c30d70411e467625b8ca618a977eb20).
Author contributions
HS collected, analyzed, and interpreted the data and prepared the draft manuscript. KF conceptualized the study and its objectives, supervised the study, and prepared the draft manuscript. DU, AY, and YM analyzed datasets. RK, TI, KK, HH, and SH extracted the data and provided clinical information. KM and KH helped in data interpretation and draft manuscript preparation. IS participated in the design and supervised the conduct of the study. All authors read and approved the final manuscript
Declaration of competing interest
The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Dr. Fujii received speaking fees from Terumo. Dr. Ueda reports grants from LPIXEL Inc. outside the submitted work. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
Acknowledgments
The authors thank Yoshikazu Hashimoto and Mitsuhiro Inomata for technical assistance regarding deep learning.
References (31)
- et al.
Assessment of culprit lesion morphology in acute myocardial infarction: ability of optical coherence tomography compared with intravascular ultrasound and coronary angioscopy
J. Am. Coll. Cardiol.
(2007) - et al.
Diagnostic accuracy of optical coherence tomography and integrated backscatter intravascular ultrasound images for tissue characterization of human coronary plaques
J. Am. Coll. Cardiol.
(2006) - et al.
Assessment of coronary plaque with optical coherence tomography and high-frequency ultrasound
Am. J. Cardiol.
(2000) - et al.
Assessment of coronary arterial plaque by optical coherence tomography
Am. J. Cardiol.
(2006) - et al.
Consensus standards for acquisition, measurement, and reporting of intravascular optical coherence tomography studies: a report from the International Working Group for Intravascular Optical Coherence Tomography Standardization and Validation
J. Am. Coll. Cardiol.
(2012) - et al.
Interobserver variability in assessments of atherosclerotic lesion type via optical frequency domain imaging
J. Cardiol.
(2021) - et al.
Accuracy of OCT, grayscale IVUS, and their combination for the diagnosis of coronary TCFA: an ex vivo validation study
JACC Cardiovasc Imaging
(2015) - et al.
Pitfalls in plaque characterization by OCT: image artifacts in native coronary arteries
JACC Cardiovasc Imaging
(2011) - et al.
Non-fibroatheroma lesion phenotype and long-term clinical outcomes: a substudy analysis from the PROSPECT study
JACC Cardiovasc Imaging
(2013) - et al.
A deep learning approach for assessment of regional wall motion abnormality from echocardiographic images
JACC Cardiovasc Imaging
(2020)