Computer Vision News - November 2020

3 Franco Matzkin 1 Best of MICCAI 2020 For both, they implemented a standard UNet , and for the first one, tested two baseline approaches: a PCA-based method for generating the bone flaps and an autoencoder method for reconstructing that flap. Franco smiles: “We tested these approaches in real decompressive craniectomy images and it turns out that our method based on a standard UNet performs better than these baseline models!” several computer vision techniques , including registration of the CT images to a common atlas, resampling to isotropic resolution, and thresholding for extracting the bone,” he explains. “Given a pre-processed CT image with decompressive craniectomy, we explored two main strategies. In the first, we predicted the full skull and then subtracted for obtaining the bone flap, and in the second strategy we directly predicted the bone flap.”