3D X-ray Image Reconstruction Based on Collaborative Neural Radiance Field and Ensemble Learning

This research examines the utilization of an integrated learner model that combines Generative Adversarial Network (GAN) and Neural Radiation Field (NeRF) for image reconstruction. The advancement of deep learning technology has led to significant progress in image reconstruction, particularly in addressing conventional issues such as image blurring, noise interference, and high computational demands. This work presents a model that integrates GRAF (Generated Radiation Field) with a unified learner designed to deduce anatomical 3D structures from a limited number or a single observation of X-rays. The model is trained using three distinct configurations of GAN models to enhance its stability and performance. In a generative adversarial network, the generator (G) strives to make counterfeit data that is indistinguishable from authentic data, whereas the discriminator (D) endeavors to differentiate between the genuine data and the counterfeit data generated by G. Through this adversarial training technique, the two entities iteratively co-evolve during the game to enhance model performance. The experimental results indicate that the model attains a peak signal-to-noise ratio (PSNR) of 35.01 and a structural similarity index (SSIM) of 0.928, demonstrating superior image quality and enhanced structural similarity. This research examines the model's potential and limitations in addressing the characteristics and issues of imaging data.