A team of researchers from Mayo Clinic in Rochester, Minn., Massachusetts General Hospital and Brigham and Women's Hospital Center for Clinical Data Science — both in Boston — and technology vendor Nvidia developed an artificial intelligence model to generate "synthetic MRIs," according to a Nvidia blog post.
A promising use for AI in healthcare is to help physicians interpret medical images. However, many healthcare organizations lack accurate and reliable imaging data needed to train these AI models. To address this issue, the research team developed a deep learning model that generates synthetic MRIs on which they can train future AI systems.
"Data diversity is critical to success when training deep learning models," the researchers wrote in a study published on the scientific preprint website arXiv.org. "Medical imaging data sets are often imbalanced as pathologic findings are generally rare, which introduces significant challenges when training deep learning models."
For the study, the researchers used a system and deep learning framework from Nvidia to train a generative adversarial network on data from two publicly available datasets of brain MRIs. They then used the network to create synthetic abnormal MRIs with brain tumors.
The researchers suggested medical institutions could share data they generate using these methods with other organizations, noting that since the images are synthetically generated, they pose no patient privacy concerns.
"This offers an automatable, low-cost source of diverse data that can be used to supplement the training set," the researchers wrote. "For example, we can alter a tumor's size, change its location or place a tumor in an otherwise healthy brain, to systematically have the image and the corresponding annotation."
To download the research study, click here.