College - Author 1
College of Engineering
Department - Author 1
Electrical Engineering Department
Degree Name - Author 1
BS in Electrical Engineering
College - Author 2
College of Engineering
Department - Author 2
Electrical Engineering Department
Degree - Author 2
BS in Electrical Engineering
Date
6-2020
Primary Advisor
Helen Yu, College of Engineering, Electrical Engineering Department
Abstract/Summary
Alzheimer’s Disease ranks (AD) as one of the most common diseases in America. Currently, detecting Alzheimer’s Disease relies upon reported symptoms, however changes in the brain can manifest years or decades before symptoms appear. In recent years, researchers have successfully utilized Artificial Neural Networks (ANN) for a variety of image classification tasks. Here, we train an ANN to detect Alzheimer’s Disease using magnetic resonance imaging (MRI) brain scans. Giving an MRI to both a neural network and a doctor will allow the doctor to be more confident in their answer, as well as double-check their answer with an objective report from a trained network.
Classifying and diagnosing Alzheimer’s disease has proven to be a problem in the past, with many cases going unnoticed until full cognitive impairment of the patient has occurred. Giving patients more time to prepare and delay the effects of Alzheimer’s and Dementia allows for a more successful and invigorating life, while ensuring the proper counter measures are being implemented as appropriate. This project specifically addresses the need for more reliable Alzheimer’s Disease detection. Because machine learning algorithms exceed human ability for detecting patterns in abstract data, neural networks could assist doctors diagnosing AD. Current detection relies primarily on user-reported symptoms. Using the trained network can provide an opportunity for early detection of Alzheimer’s Disease, enabling those preventative measures.
In this project, we designed a convolutional neural network to detect Alzheimer's Disease from 3-D MRI images. To do so, we slice each 3-D scan into multiple 2-D images and input each one to the network individually. Using an adapted InceptionV3 architecture, we achieve an testing accuracy of about 72% on individual 2-D image slices. We then post-process the results by averaging the classification scores over each slice of the MRI scan, which increases the classification accuracy to just over 88%.
URL: https://digitalcommons.calpoly.edu/eesp/509