Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Deep Learning-Based Co-Registration of Coronary Computed Tomography and Intravascular Images
Coronary atherosclerosis can be assessed with a wide array of imaging tools, each with its own strengths in diagnosis and treatment planning. However, aligning such imaging modalities is difficult and time-consuming. This project aims to develop a deep learning tool that can achieve multi-modal co-registration of computed tomography and intravascular images of coronary artery disease. Such a tool can be used to rapidly and automatically align large multi-modal imaging datasets, paving the way for various clinical and machine learning applications.
Keywords: Deep Learning, Co-registration, Medical Imaging, Cardiology, Computer Vision, Computational Geometry
Coronary Computed Tomography Angiography (CCTA) is a 3-dimensional imaging modality that offers crucial information on the presence, extent, and severity of obstructive coronary artery disease (CAD). Patients undergoing CCTA typically receive a contrast dye injection, enabling visualization of coronary anatomy. Although CCTA is widely used, it primarily focuses on luminal assessment, with limited capabilities for evaluating soft tissue intraplaque components and distinctive blooming artifacts in the presence of intraplaque calcium deposits.
Multiple studies have quantified CCTA's effectiveness in assessing CAD-related diagnostic metrics, such as luminal area, calcium morphology, and plaque burden. Most of these studies validate CCTA's performance by co-registering image slices with invasive intravascular imaging modalities like Optical Coherence Tomography (OCT), which offers higher fidelity visualization of the lumen and surrounding diseased tissue. We have recently developed a semi-automatic pipeline that aligns intravascular image frames along the artery to their equivalent frames in CCTA images. This project aims to integrate deep-learning modules into the pipeline to automatically align OCT and CCTA images. The resulting module will serve as a foundation for further research projects examining the relationship between calcium micromorphology, arterial biomechanics, clinical intervention success rates, and major cardiovascular events.
Coronary Computed Tomography Angiography (CCTA) is a 3-dimensional imaging modality that offers crucial information on the presence, extent, and severity of obstructive coronary artery disease (CAD). Patients undergoing CCTA typically receive a contrast dye injection, enabling visualization of coronary anatomy. Although CCTA is widely used, it primarily focuses on luminal assessment, with limited capabilities for evaluating soft tissue intraplaque components and distinctive blooming artifacts in the presence of intraplaque calcium deposits. Multiple studies have quantified CCTA's effectiveness in assessing CAD-related diagnostic metrics, such as luminal area, calcium morphology, and plaque burden. Most of these studies validate CCTA's performance by co-registering image slices with invasive intravascular imaging modalities like Optical Coherence Tomography (OCT), which offers higher fidelity visualization of the lumen and surrounding diseased tissue. We have recently developed a semi-automatic pipeline that aligns intravascular image frames along the artery to their equivalent frames in CCTA images. This project aims to integrate deep-learning modules into the pipeline to automatically align OCT and CCTA images. The resulting module will serve as a foundation for further research projects examining the relationship between calcium micromorphology, arterial biomechanics, clinical intervention success rates, and major cardiovascular events.
• Developing a co-registered dataset of CT-OCT images to train machine learning algorithms.
• Adapting and improving a previously developed spatial transformation module for optimization-based alignment of CT and OCT image pairs.
• Leveraging recent advances in deep learning and image registration to automatically align CT-OCT image pairs.
• Developing a co-registered dataset of CT-OCT images to train machine learning algorithms.
• Adapting and improving a previously developed spatial transformation module for optimization-based alignment of CT and OCT image pairs.
• Leveraging recent advances in deep learning and image registration to automatically align CT-OCT image pairs.
Interested candidates should send CV and cover letter to kkadry@mit.edu
Interested candidates should send CV and cover letter to kkadry@mit.edu
