A New Informatics Approach for Detection of Cerebrovascular Abnormalities - The goal of this clinical informatics project is to develop computational techniques to model and analyze brain blood vessels for detecting morphometric abnormalities that are hallmarks of cerebrovascular diseases (CVDs). The project addresses an important challenge in neuroradiology and neurosurgery: how to accurately diagnose CVDs on computed tomography angiography (CTA). CVDs include intracranial aneurysms, stroke, intracranial vascular stenosis, dural fistula, and other disorders of the brain vasculature, and these diseases have severe outcomes as they cause hemorrhage, stroke, neurological damage, and death. In fact, each year, CVDs cause more than 100,000 deaths in the US, and an even larger population suffers permanent damage, including stroke, paralysis, and loss of speech. If we can diagnose CVDs more accurately and promptly, mortality and morbidity can be significantly reduced. Brain imaging is a first line diagnostic for CVDs with the image hallmarks being brain blood vessel abnormalities. Yet diagnosis is very challenging because a clinician needs to sift through and zoom in and out of and rotate a large number of images to examine each blood vessel for malformation, whether it is a narrowing or the formation of intracranial aneurysms on blood vessel walls. Similarly, a neurosurgeon needs to read brain scans right before an operation to locate the positions of abnormalities. Our specific aims of this project are to develop novel computational techniques including deep learning to model and analyze blood vessels to detect abnormalities and highlight their locations for clinicians to examine further. While computers are not yet sophisticated enough to make diagnoses like a trained clinician, computers can perform more objectively and quickly, compared to human experts, the necessary complex shape analysis and quantification, such as identifying abnormal widening or narrowing of blood vessels and detecting protrusions on blood vessel walls. To address the request from clinicians that they would benefit significantly from computer-aided detection of abnormalities and, once abnormalities are marked, they can make highly accurate diagnosis and classification of the underlying CVDs, we designed an informatics approach as a computer-aided tool to analyze CTA images. We will model both individual blood vessels and the whole vasculature in the 3D space. Then, from the vasculature, we will develop and implement a multi- channel deep learning model focused on shape analysis to detect blood vessel abnormalities. Finally, abnormalities will be marked in colors in 3D to allow clinicians to make more accurate diagnoses, plan preventative treatments, and perform precise surgeries to benefit patient health.
