Fig. 6
From: Deep learning-based metal artifact reduction in CT for total knee arthroplasty
The left knee joint of a 78-year-old woman who underwent revision total knee arthroplasty surgery due to aseptic loosening of the femoral component. (a) A preoperative lateral knee radiograph taken before revision surgery shows radiolucent gaps around the anterior and posterior flanges of the femoral component (long arrows). Axial noncontrast CT images reconstructed with (b) Non-MAR, (c) O-MAR protocol, and (d) KMAR-Net are shown in the bone window setting (window width = 2000 HU, window level = 500 HU). The image of the Non-MAR protocol is of non-diagnostic image quality due to severe streak artifacts. In the O-MAR image, new artifacts of high attenuation that are not visible in the original image interfere with the evaluation of cortical and trabecular bones (arrowhead). KMAR-Net rarely shows these hyperattenuating artifacts while further reducing streak artifacts, and the bone-implant interface gap is more clearly demonstrated (arrows). MAR, metal artifact reduction; O-MAR, metal artifact reduction algorithm for orthopedic implants; KMAR-Net, knee metal artifact reduction network.
