Implant Segmentation in Radiographic Imagery Using Wavelet Decomposition and Multiresolution MTANN

Occasionally implants are covered with a musculoskeletal system in medical imaging, it could be difficult for medical personnels as well as computers to perceive these implants. In this paper, we use image processing techniques to suppress the musculoskeletal system contrast in X-ray images that contain implants through artificial neural networks called massive training artificial neural networks (MTANN). MTANN is a non-linear filter that can be trained using appropriate X-ray images input and learning images. Learning images are obtained through the editing process using a graphical editor, which is used to correct learning images from noise. To effectively suppress a musculoskeletal system that has multiple spatial frequencies, we use multiresolution MTANN consisting of wavelet decomposition/composition techniques and dedicated MTANN for every resolution. Multiresolution MTANN hopefully can provide images that are really close to learning images. By reducing the contrast of the musculoskeletal system from X-ray images, we will be able to create X-ray images where the musculoskeletal system is substantially suppressed and be able see the implant clearly in contrast. We use MURA as a training database and for validation tests. MURA is a large set of bone radiographic data, which is designed to train ANN to determine whether X-ray images are normal or abnormal. Once the model is done, when our technique is applied to non-training X-ray images, the musculoskeletal image system is substantially suppressed while maintaining the visibility of the implant. Our image processing technique to suppress musculoskeletal systems using multiresolution MTANN will potentially be useful to medical personnels.