Document Type
Thesis
First Faculty Advisor
Blais, Brian
Keywords
Deep Learning; Computer Vision; Neural Networks
Publisher
Bryant University
Rights Management
All rights retained by Bryant University and Anthony Pasquarelli
Abstract
This project explores the use of deep learning to predict age based on pediatric hand X-Rays. Data from the Radiological Society of North America’s pediatric bone age challenge were used to train and evaluate a convolutional neural network. The project used InceptionV3, a CNN developed by Google, that was pre-trained on ImageNet, a popular online image dataset. Our fine-tuned version of InceptionV3 yielded an average error of less than 10 months between predicted and actual age. This project shows the effectiveness of deep learning in analyzing medical images and the potential for even greater improvements in the future. In addition to the technological and potential clinical benefits of these methods, this project will serve as a useful pedagogical tool for introducing the challenges and applications of deep learning to the Bryant community.
Included in
Databases and Information Systems Commons, Numerical Analysis and Scientific Computing Commons
