Computer Vision News - May 2018

Min is presenting her paper which explores deep learning methods for medical image analysis. She uses a segmentation-by-detection method , which means that rather than segmenting an original image directly, it uses a detection module first to detect the region of interest. With a narrower search area, it can focus the most algorithms on the region of interest to segment the detail of the object. This method is much more efficient as it cuts down the time it takes to search directly on the volume. Accuracy is improved too, as by narrowing down the area of interest you can get rid of the surrounding noise, which is particularly important for medical data. Min explains that the challenging part is how to combine the detection method with the segmentation method . Also, if you already have the detection region, how to combine it to the 3D segmentation using the deep neural networks. 30 Saturday ISBI DAILY Poster Presentation: Segmentation-By-Detection: A Cascade Network for Volumetric Medical Image Segmentation Min Tang Min Tang is a PhD student at the University of Alberta, supervised by Martin Jagersand and Dana Cobzas . She spoke to us ahead of her poster session at ISBI 2018. “ If you want to make it more accurate, instead of using the detection part you can use a human input .” BEST OF