2022/03/17 by Fuxin Fan, Ludwig Ritschl, Fan, Fuxin +13
Engineering · Medicine · Physics and Astronomy · #Advanced X-ray Imaging Techniques #Advanced X-ray and CT Imaging #Artificial intelligence #Artificial neural network #Computer Vision and Pattern Recognition (cs.CV) #Computer science #Computer vision #Convolutional neural network #Encoder #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Medical Imaging Techniques and Applications #Mineral Processing and Grinding #Pattern recognition (psychology) #Robustness (evolution) #Segmentation #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2203.09207
openalex publication_date 2022/03/17 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
In several image acquisition and processing steps of X-ray radiography,\nknowledge of the existence of metal implants and their exact position is highly\nbeneficial (e.g. dose regulation, image contrast adjustment). Another\napplication which would benefit from an accurate metal segmentation is cone\nbeam computed tomography (CBCT) which is based on 2D X-ray projections. Due to\nthe high attenuation of metals, severe artifacts occur in the 3D X-ray\nacquisitions. The metal segmentation in CBCT projections usually serves as a\nprerequisite for metal artifact avoidance and reduction algorithms. Since the\ngeneration of high quality clinical training is a constant challenge, this\nstudy proposes to generate simulated X-ray images based on CT data sets\ncombined with self-designed computer aided design (CAD) implants and make use\nof convolutional neural network (CNN) and vision transformer (ViT) for metal\nsegmentation. Model test is performed on accurately labeled X-ray test datasets\nobtained from specimen scans. The CNN encoder-based network like U-Net has\nlimited performance on cadaver test data with an average dice score below 0.30,\nwhile the metal segmentation transformer with dual decoder (MST-DD) shows high\nrobustness and generalization on the segmentation task, with an average dice\nscore of 0.90. Our study indicates that the CAD model-based data generation has\nhigh flexibility and could be a way to overcome the problem of shortage in\nclinical data sampling and labelling. Furthermore, the MST-DD approach\ngenerates a more reliable neural network in case of training on simulated data.\n