Automatic Spatiotemporal Analysis of Cardiac Image Series

Cardiovascular disease continues to be the leading cause of death in North America. In adult and, alarmingly, ever younger populations, the so-called obesity epidemic largely driven by lifestyle factors that include poor diet, lack of exercise and smoking, incurs enormous stresses on the healthcare system. The primary cause of serious morbidity and mortality for these patients is atherosclerosis, the build up of plaque inside high pressure vessels like the coronary arteries. These lesions can lead to ischemic disease and may progress to precarious blood flow blockage or thrombosis, often with infarction or other severe consequences. Besides the stenosis-related outcomes, the arterial walls of plaque-ridden regions manifest increased stiffness, which may exacerbate negative patient prognosis. In pediatric populations, the most prevalent acquired cardiovascular pathology is Kawasaki disease. This acute vasculitis may affect the structural integrity of coronary artery walls and progress to aneurysmal lesions. These can hinder the blood flow's hemodynamics, leading to inadequate downstream perfusion, and may activate thrombus formation which may lead to precarious prognosis. Diagnosing these two prominent coronary artery diseases is traditionally performed using fluoroscopic angiography. Several hundred serial x-ray projections are acquired during selective arterial infusion of a radiodense contrast agent, which reveals the vessels' luminal area and possible pathological lesions. The acquired series contain highly dynamic information on voluntary and involuntary patient movement: respiration, organ displacement and heartbeat, for example. Current clinical analysis is largely limited to a single angiographic image where geometrical measures will be performed manually or semi-automatically by a radiological technician. Although widely used around the world and generally considered the gold-standard diagnosis tool for many vascular diseases, the two-dimensional nature of this imaging modality is limiting in terms of specifying the geometry of various pathological regions. Indeed, the 3D structures of stenotic or aneurysmal lesions may not be fully appreciated in 2D because their observable features are dependent on the angular configuration of the imaging gantry. Furthermore, the presence of lesions in the coronary arteries may not reflect the true health of the myocardium, as natural compensatory mechanisms may obviate the need for further intervention. In light of this, cardiac magnetic resonance perfusion imaging is increasingly gaining attention and clinical implementation, as it offers a direct assessment of myocardial tissue viability following infarction or suspected coronary artery disease. This type of modality is plagued, however, by motion similar to that present in fluoroscopic imaging. This issue predisposes clinicians to laborious manual intervention in order to align anatomical structures in sequential perfusion frames, thus hindering automation of