Abstract

MRIs are a powerful imaging tool to help diagnose and treat complex conditions. Since its widespread adoption in the late 20th century, healthcare organizations have grown more reliant on its ability to guide physician’s decision making. This has led to these institutions becoming inundated with big medical imaging files that require large amounts of data storage and lead to large operational costs. To some degree, the introduction of cloud storage has helped offset some of these costs and provide flexibility to this organizations, but there still exists a need for compression methods to further alleviate the burden of this big data problem. To this end, in this work, we introduce a method to compress MRI images using Singular Value Decomposition (SVD) and benchmark the performance of a pre-trained Convolutional Neural Network (CNN) when trained on the uncompressed images as compared to compressed images for a diagnostic task.

Full Paper