Artificial intelligence holds a great deal of promise for medical professionals who want to get the most out of medical imaging. However, when it comes to studying brain tumors, there's an inherent problem with the data: abnormal brain images are, by definition, uncommon. New research from Nvidia aims to solve that.
A group of researchers from Nvidia, the Mayo Clinic, and the MGH & BWH Center for Clinical Data Science this weekend are presenting a paper on their work using generative adversarial networks (GANs) to create synthetic brain MRI images. GANs are effectively two AI systems that are pitted against each other -- one that creates synthetic results within a category, and one that identifies the fake results. Working against each other, they both improve.
GANs could help expand the data sets that doctors and researchers have to work with, especially when it comes to particularly rare brain diseases.
"Diversity is critical to success when training neural networks, but medical imaging data is usually imbalanced," Hoo Chang Shin, a senior research scientist at Nvidia, explained to ZDNet. "There are so many more normal cases than abnormal cases, when abnormal cases are what we care about, to try to detect and diagnose."
Also: Top 5: Things to know about AI TechRepublic
Shin and others are presenting their research at the MICCAI conference in Spain, which explores the intersection of computer science and medical imaging.
In addition to widening the potential data sets, Shin and his colleagues say using GANs could provide a solution for the privacy challenges that surround the use of patient data. Because the synthetic images are not tied to a specific patient, it's more anonymous and safer to transfer outside of a hospital.
The research team used an Nvidia DGX-system with the cuDNN-accelerated PyTorch deep learning framework to train the GAN on data from two publicly available data sets of brain MRIs -- one with images of brains with Alzheimer's disease, and the other with images of brains with tumors.
The GAN was trained with a brain anatomy label and a tumor label separately, meaning the team can alter either the tumor label or the brain label produce synthetic images with desired characteristics -- such as a tumor of a certain size or location in the brain.
However, Shin explained, because the biology of the tumor is not entirely understood, the team can't just create an image of a tumor from scratch -- the GAN needs to start with at least one real image of a tumor.
To advance this research, Shin said blind testing should be conducted to ensure the quality of the synthetic images. Additionally, more work should be done to ensure the privacy of patients from the original data sets is indeed protected. Ultimately, the goal is for GAN imaging to help doctors learn more about rare brain tumors.
Previous and related coverage:
An executive guide to artificial intelligence, from machine learning and general AI to neural networks.
The lowdown on deep learning: from how it relates to the wider field of machine learning through to how to get started with it.
This guide explains what machine learning is, how it is related to artificial intelligence, how it works and why it matters.
An introduction to cloud computing right from the basics up to IaaS and PaaS, hybrid, public, and private cloud.
It is a capital improvement project the size of the entire planet, replacing one wireless architecture created this century with another one that aims to lower energy consumption and maintenance costs. Sure, you'll download movies faster on your phone, but that's not the real reason 5G's arrival is being accelerated.