Cardio-respiratory motion compensation for radiation dose reduction in X-ray guided cardiac interventions

Abstract

Cardiac diseases affect a large population and especially children, every year. Congenital Heart Disease (CHD) is the most common type of birth malformation worldwide. CHD is caused by abnormalities in heart structure before birth. It is imperative to advance the treatment techniques to be as less invasive as possible. One popular treatment for CHD is X-ray image-guided interventional procedures that have gained popularity over the past two decades. Navigation guidance during cardiac interventions, such as balloon angioplasty and stent placement, is generally performed under X-ray fluoroscopy. Patients with CHD are exposed to substantial amounts of ionizing radiation from diagnostic and treatment procedures. In recent years, the number of complex, long-duration pediatric cardiac interventions has risen significantly. During the cardiac intervention, several organs, including the arteries, are moving, given the heart beating, respiratory movement, and sometimes the patient’s movements. These movements degrade image acquisition and make the navigation process more difficult. This research’s main goal was to develop less invasive techniques to apply for pediatric cardiac interventions. We pursued this goal by minimizing the radiation dose the patient and staff received and compensating for the induced motions by estimating and predicting the targets’ (arteries) movements. Moreover, while the targets’ movements are tracked in the images, the need to inject the contrast agent to visualize the vessels will be reduced. In the context of this research, we developed and validated our approaches using both simulated and patient X-ray angiography datasets from Sainte-Justine Hospital. Simulated X-ray sequences generated from realistic XCAT computational phantoms with cardio-respiratory motion were first investigated. The simulated motion included the beating heart and respiratory motion. We simulated 56 different patients (32 male and 24 female) and 112 sequences (2 sequences per patient, showing either the left or the right coronary artery). All the generated sequences had a length of 75 frames and were generated at 15 frames per second (fps). The patient X-ray angiography dataset comprises 52 different patients with contrasted coronary arteries. Each patient presents a different number of sequences with varying lengths. There is a total number of 340 sequences, with a minimum and maximum length of 15 and 70 frames, respectively. All the data were acquired at 15 fps. In the first objective, a generative learning-based approach was proposed to predict X-ray angiography frames to reduce the amount of radiation exposure to pediatric patients and the staff during cardiac interventions while preserving the image quality. In the second objective, we focused on extracting 2D motion features from the X-ray sequences first and then building up a "predict-ahead" motion model for navigating the interventions. Our model-free cardio-respiratory motion estimation approach can predict cyclic cardio-respiratory motion signals artifacted by sudden motions caused by the patients or some irregularities. This approach was developed and validated with both simulated and patient datasets. For the third objective, using the simulated dataset and based on our experiment on the second objective, we investigated the different motion patterns between males and females. The research will offer a general solution that does not require additional imaging modality while providing an accurate motion prediction and estimation. The methodologies to achieve these objectives are based on deep learning algorithms by first extracting motion features from the images and tracking these features. We believe that learning-based approaches can pave the road for better assessment of cardiovascular motion and radiation dose reduction for the patients and staff. Thus, in this thesis, we have applied deep learning methods to facilitate the desired less-invasive cardiac interventions.

Date	22 Nov 2022
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Luc Duong (Supervisor)