A Hybrid Generative Adversarial network with Quantum U-NET for 3D spine X-ray image registration

Abstract

The registration of Two Dimensional (2D) to Three Dimensional (3D) spine x-ray image is essential in medical field for the diagnosis of scoliosis. In the surgerical procedure of scoliosis, Computed tomography (CT) imagery and intraoperative projective X-rays are used with the requirement of implant localization. The 3D reconstruction is challenged with the presence of implant projection which affects the registration capture range. In this paper, the novel inpainting algorithm Generative Adversarial Network with Quantum Unet (GAN-Q-Unet) model is developed to resolve the issues of registration capture range. The generator network learns to produce data with the output of discriminator network. In GAN-Q-Unet, the discriminator network is replaced with the encoder-decoder network namely U-net which improves the quality of the network by providing more semantic information. In addition to that, the layers of U-net are replaced with quantum layers in which encoding and transformation is applied based on non-linearity. Initially, the bone regions of the spine image are enhanced with synthetic data generation. After enhancement, the proposed inpainting is applied for 3D image registration. With GAN-Q-Unet approach, better 3D reconstruction of spine image is obtained from 2D spine image. The proposed algorithm is used for automatic transformation of spine image with varying dimensions in the routine examination of clinical applications. By using the proposed approach, the Peak Signal to Noise Ratio (PSNR), Structural Similarity index Measure (SSIM), Dice Similarity coefficient (DSC) and Loss values obtained are 28.1767, 0.9300, 2.9674, and 0.9107 respectively. The registration performance of GAN-Q-Unet is compared with the state-of-the-art techniques and it shows the efficiency of the proposed 3D image registration.