Deep Learning Image Analysis Model Of Pulmonary Arteries

The use of artificial intelligence in cardiovascular image analysis has become an increasingly useful tool, due to the ability of accurately sharpening images of heart and blood vessel detailed anatomy. Deep learning, a subset of artificial intelligence, has been proposed as an efficient platform in screening for coronary artery disease, pulmonary embolism, and to distinguish between pulmonary arteries and veins. In the case of coronary artery disease, lumen delineation proves difficult, though arteries can be identified more accurately with the use of deep learning. With this, patients can be screened for coronary artery disease prior to invasive coronary angiographies. Such data science techniques can also be used for other diseases that are difficult to diagnose such as pulmonary artery disease. For example, the gold standard in diagnosing pulmonary arterial hypertension (PAH) is invasive right heart catheterization. During the progression of PAH, the pulmonary arteries undergo a remodeling process which leads to pruning, or occlusion, of the smaller arteries and increasing diameter and stiffness of the larger arteries. Current treatments alleviate symptoms, but do not cure PAH. Also, due to the non-specific symptoms of PAH, many cases go undiagnosed for quite some time. This work explores the use of a convolution neural network (supervised), multi-scale cluster analysis (unsupervised), and a hybrid deep learning approaches on images from the public lung image database as a potential tool for pre-screening pulmonary arterial disease.