Computer Vision News - August 2020

Kaiyang Cheng (Victor) is an undergraduate student at the University of California, Berkeley. He is also co-first author on a novel paper that has just won the Best Paper award at this year’s MIDL conference. Victor has been interning in the lab of co-author Valentina Pedoia, an assistant professor in the radiology department at the University of California, San Francisco. Together with co-first author Francesco Calivà, a postdoc at the Center for Intelligent Imaging in Valentina’s lab, they speak to us about the background to their prize-winning research. Over the past few years, deep learning models have been shown to be successful in accelerating MRI reconstruction . MRI reconstruction can be slow when trying to collect the entire k-space data. To accelerate this, you can collect a subset of the k-space data when acquiring the images. However, this can degrade the image quality and lead to a loss of diagnostic detail. The concept here is the same as imagecompression . When you compress an image, you accept the fact that you sacrifice some detail. You might lose a little bit of an ear of a cat, but it is not a big deal. However, with medical images, losing even a very small percentage of the entire image means you risk losing important diagnostic information , such as abnormalities in the ground truthpatient’s anatomy. If one pixel is lost and the abnormality is within that pixel, the image Addressing the False Negative Problem of Deep Learning MRI Reconstruction Models by Adversarial Attacks and Robust Training Best Paper Award 12 Best of MIDL 2020 Best Paper Award