Researchers are using artificial intelligence to create high-quality medical images which expose patients to less radiation.
Radiologists require high-quality images based on large amounts of data to make accurate diagnoses but acquiring sufficient data means exposing patients to large doses of radiation in CT scans or uncomfortably long MRI scans.
The team of researchers, from the Martinos Center for Biomedical Imaging at Massachusetts General Hospital (MGH), have trained a neural network which is able to reconstruct an image based on less data, meaning radiologists could get the same results with lower radiation doses or shorter scans.
Their technique, called automated transform by manifold approximation (AUTOMAP), involves training a neural network to understand what makes an image an image.
With this prior knowledge in place, it should be able to more quickly encode an image generated via a sensor during the medical scan.
Writing in the journal Nature, the MGH team said their research, “recasts image reconstruction as a data-driven supervised learning task that allows a mapping between the sensor and the image domain to emerge from an appropriate corpus of training data.”
This marks an important breakthrough in medical imaging, capitalizing on recent advances in neural networks, the speed of graphical processing units (GPU), and the potential of big data.
Image reconstruction typically requires enormous computation power, but neural networks once trained, can run on inexpensive GPUs.
Matthew Rosen, co-director of the Center for Machine Learning at MGH’s Martinos Center said, “We have a Clinical Data Science Center which enables us to push our algorithms to the terminals of the radiologists.
“In the next 6 months, maybe, we’ll start doing that. We’ll have some studies where our reconstructions are running side-by-side with the state-of-the-art, FDA-approved algorithms and let radiologists give us feedback on what they think.”
For a Deep Dive on medical imaging please see the third issue of the AIMed Magazine here.