Characterization of coronary artery tissues with Kawasaki disease from OCT images

Abstract

Kawasaki Disease (KD), mucocutaneous lymph node syndrome, is an acute childhood vasculitis syndrome, which is characterized by fever, rash, bilateral nonexudative conjunctivitis, erythema of the lips and oral mucosa, and swollen erythematous hands and feet. KD is an inflammatory disease, which leads to inflammation in the walls of medium-sized arteries throughout the body. Although a high dose of Intravenous Immune Globulin (IVIG) infusion reduces the risk of coronary artery complications, about 15% to 25% of untreated children suffer a risk of experiencing coronary artery aneurysms or ectasia. Intimal thickening, media disappearance, lamellar calcifications, fibrosis, macrophage, and neovascularization are the most distinguished pathological features of late coronary artery lesions in Kawasaki disease. In severe cases, they can lead to myocardial infarction and sudden death. Since the functionality of the cardiac tissues significantly depends on the coronary blood flow to the myocardium, intravascular assessment of coronary artery tissues is significant to detect the pathological formations caused by different coronary artery complications. Although in vivo intravascular visualization of coronary arteries is significant to provide highly valuable progressive information, it is a challenging task, especially in pediatric patients because of the small size of the vessels and high heart rate. OCT is an interferometric imaging modality that maps the backscattered near-infrared (NIR) light to create cross-sectional images of the tissues under review in micrometer scale. OCT was developed for the diagnosis and treatment guidance of coronary artery disease in the adult population. It has been recently used in pediatric cardiology to image coronary artery tissues with appropriate safety results in this age group. It has high resolution ranging from 10 to 20 μm to characterize the internal structure of the tissues such as vessel wall layers and plaque accumulation. Inner vessel wall geometry allows detecting and evaluating biophysical and dynamic properties of arterial wall, the thickness of coronary artery layers, and various coronary artery abnormalities caused by the disease. This thesis is focused on developing an intra-coronary tissue characterization model using OCT imaging to pave the way for evaluating the functionality of coronary artery tissues. The experiments are performed on intracoronary OCT acquisitions from patients affected by Kawasaki disease. Analysis of coronary artery tissues is a broad study field, which consists of three main steps: 1. Classification of coronary artery layers to recognize characteristic attributes of each layer, intima and, media. 2. Identification of coronary artery lesions caused by KD on coronary artery tissues to assess the functionality of coronary arteries. 3. Motion correction as the step of 3D reconstruction for longitudinal and transversal assessment of different pathological formations and estimation of arterial wall stiffness. For the first contribution, we developed an automatic classification approach to characterize coronary artery layers in pediatric patients using the images obtained from OCT system. The goal of the study was to identify the features, which perfectly describe intima and media layers using a Convolutional Neural Network (CNN). The activations of the last fully connected layer are used to train Random Forest (RF) for the classification task. This work contributes to evaluating the thickness of coronary artery layers to distinguish between normal and diseased segments of the coronary artery. A motion correction model of intracoronary OCT images is proposed for the second contribution. Our algorithm is designed for intra-slice motion correction in intravascular OCT images using tissue information rather than the lumen outline. Features are extracted automatically by applying a Convolutional Neural Network and the similarity between deep features is used to perform registration. For the first time, deep learning is applied on intracoronary OCT images for motion correction. This will contribute to evaluate the functionality of coronary arteries by analyzing the volume variation and considering the motion of the vessel. Also, it is a robust method to assess the pathological formations by finding the correlation between the tissues of adjacent frames. For the third contribution, we focused on developing a tissue characterization approach to classifying various pathological formations of coronary arteries caused by KD. Specifically, the most distinguished coronary artery complications such as fibrosis, macrophage, neovascularization, and calcification as well as coronary artery layers (intima, and media) are detected using deep features extracted from pre-trained CNNs and majority voting from Random Forest classification. This study contributes to preventing future complications in children and young adults suffered from Kawasaki disease. Since mentioned pathological formations are recognized as the most common intracoronary complications caused by coronary artery disease (CAD), this work is not limited to intracoronary tissue characterization in KD patients.

Date	27 Jun 2018
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Luc Duong (Supervisor) & Nagib Dahdah (Co-supervisor)