Using deep learning to remove electromagnetic interference in MRI scanners
MRI scans are a very useful clinical diagnostic tool. They give high fidelity, 3D images of the inside of the body. But, they are large, expensive machines and there are not enough of them to go around, particularly in low income countries.
Significant amounts of work has gone into finding a lower cost alternative to MRIs that requires less shielding and lower power. A low cost MRI scanner could have a huge impact on health outcomes all over the world. So far, all attempts to build such a device have not shown sufficient image quality.
One of the causes of poor image quality has been the presence of external electromagnetic interference. A new approach to dealing with this, using machine learning, was recently published in Nature. The paper describes using a convolutional neural network (CNN) to predict and remove the inteference pattern. By accurately predicting the shape of the interface that would be present in them MRI scan, it was then possible to simply subtract this from the final image.