Computer Vision News - November 2020

2 Best Paper Award (challenge) 30 Best of MICCAI 2020 Franco Matzkin is a second-year PhD student at the Research Institute for Signals, Systems and Computational Intelligence, located in Santa Fe, Argentina. His advisor is Enzo Ferrante. He speaks to us about this work on self-supervised skull reconstruction, whichwas presented at this year’s virtual MICCAI and won the Best Paper Award at the AutoImplant Challenge. For this work, Franco and his co- authors studied a set of methods for performing digital skull reconstruction in brain CT images of patients with decompressive craniectomy , which is an intervention performed after lesions such as traumatic brain injury. Solving this problem may help to design cranioplasty implants or to obtain indicators such as the intracranial volume or extracted bone flap volume . Previous works on this topic have tended to use a lengthy manual approach called ABC , where the volume of the bone flap is calculated by a doctor. Other digital reconstruction methods have needed to leverage the symmetry of the images. “That is no good because if you have a craniectomy in the front of the patient, you cannot take the symmetry and use it for generating the implant,” he points out. “We wanted a novel automatic learning process using deep learning .” To achieve this, the team implemented a self-supervised convolutional neural network model , simulating the craniectomies digitally and then testing the method in a real scenario. “We used a pre-processing step and Self-supervised Skull Reconstruction in Brain CT images with Decompressive Craniectomy