14 Poster Presentation DA I L Y Jingyu Liu is a postdoc at Peking University and a researcher at Deepwise in China. He speaks to us ahead of his poster this afternoon about chest X-ray. Jingyu’s work explores disease classification and localization in chest X-ray. He tells us that this is a weakly supervised problem setting. His contribution is to find the difference between positive and negative images. To find the contrast, you have to align the images to a target image first. How do you get that target image? Jingyu explains: “ Our method is to get a canonical image which is averaged from 500 negative chest X-rays . We use a technique borrowed from style transfer and perceptual loss and affine- transform the positive image and negative image to be upright in the last canonical image. After that, we input both positive and negative images together on feature maps and make a subtraction at the feature map to get an attention map. Then we add that attention map to the original aligned positive images to localize and classify. Align, Attend and Locate: Chest X-Ray Diagnosis via Contrast Induced Attention Network With Limited Supervision "We use a technique borrowed from style transfer and perceptual loss and affine-transform the positive image and negative image to be upright in the last canonical image."