Computer Vision News - November 2020

234 Paper Presentation (workshop) Ilya Kovler is CTO at RSIP Vision . He is presenting a paper at Thursday’s Machine Learning for Medical Image Reconstruction (MLMIR) workshop at MICCAI about his lab’s work on 3D reconstruction from X-ray images. He tells us about this and other work supporting AI for orthopedic surgery. RSIP Vision specialize in artificial intelligence for medical imaging , having worked for several years with state- of-the-art technologies and computer vision techniques, from traditional computer vision to deep learning with convolutional neural networks . Some of the company’s strongest work is in the field of AI for orthopedics where they have developed many different solutions, including bone and joint segmentation, osteophyte segmentation, and anatomical landmark detection for hip and knee joints. They also work in the intraoperative field by providing calibration capabilities and 2D/3D registration. Their most recent breakthrough is in knee segmentation and landmark detection from X-ray images , which is an extremely useful orthopedic tool allowing fully automatic pre-op planning of the implant . X-ray is one of the most common imaging modalities for the planning of joint replacement procedures. It is widely available, low cost and low radiation, making it perfect for mass-scale rollout. The challenging part is that X-ray images provide only 2D information and each pixel in the output can belong to a couple of bones. For example, the patella usually totally overlaps with the femur. However, using RSIP Vision’s approach, it is possible to create an accurate 3D model of the joint from this 2D information alone. End-To-End Convolutional Neural Network for 3D Reconstruction of Knee Bones From Bi-Planar X-Ray Images Ilya Kovler Best of MICCAI 2020