Date of Award

6-10-2021

Document Type

Thesis

Publisher

Santa Clara : Santa Clara University, 2021.

Departments

Bioengineering; Computer Science and Engineering

First Advisor

David Anastasiu

Second Advisor

Yuling Yan

Abstract

A brain aneurysm is a thin or weak spot on a blood vessel wall that expands and fills with blood. Brain aneurysms are very dangerous due to the fact that in most cases, patients do not show any symptoms. Because of this, aneurysms are difficult to diagnose unless it becomes very large or ruptures, resulting in fatal hemorrhage.

Aneurysms can be detected by a number of different brain imaging methods including Magnetic Resonance Imaging (MRI), Magnetic Resonance Angiography (MRA), Computed Tomography Angiography (CTA) and other imaging methods but for the sake of this report we will only be focusing on MRA. MRA scans are optimal for the detection of brain aneurysms because they produce images that can be used to distinguish blood vessels from surrounding stationary tissue. Since aneurysms happen only in the blood vessels, MRA scans are an ideal image type to train a predictive machine learning model for our purposes.

Artificial intelligence, machine learning, and deep learning are all growing fields that are leading to breakthroughs in the medical community. For machine learning models and algorithms specifically, they have greatly helped in analyzing, locating, and predicting critical health conditions including brain aneurysms.

With the help of this technology, medical professionals, such as radiologists, can greatly benefit from the predictions these models can provide. Image interpretation by human experts can be limited due to subjectivity, complexity of the image, extensive variations across different interpreters, and fatigue. With the help of these models, physicians and radiologists will be able to make more accurate and precise diagnoses with predictive algorithms serving as a second opinion. More specifically, with the help of machine learning and MR Angiography, a predictive model can be trained to help detect brain aneurysms that might otherwise go unnoticed by radiologists. Here, we have developed four convolutional neural network (CNN) models that successfully detect aneurysm presence within MRA scans. Further investigation should be done to validate and improve these models to create a more accurate and sensitive diagnostic platform.

COinS