NNTV-GS: a fast reconstruction method combining nearest neighbor total variation and Gaussian splatting in dental offset detector CBCT

<i>Objective.</i>This study aims to address the slow reconstruction speed of iterative reconstruction algorithms in dental cone-beam computed tomography (CBCT) imaging while enhancing reconstruction accuracy. Specifically, the objective is to introduce a high-accuracy, fast iterative reconstruction algorithm based on 3D Gaussian splatting (3DGS) that integrates truncated-artifact correction and nearest-neighbor total variation (NN-TV) to improve clinical applicability.<i>Approach.</i>A 3DGS-based iterative reconstruction algorithm is proposed, which incorporates a truncated-artifact correction mechanism and an NN-TV algorithm. The 3DGS method utilizes explicit scene representation and a differentiable rendering process, enabling direct optimization of the dental three-dimensional structure through gradient descent. The truncated-artifact correction technique mitigates artifacts caused by offset detectors commonly used in dental CBCT. Additionally, the NN-TV algorithm is integrated into the Gaussian scene representation to enhance the quality of reconstructed images.<i>Main Results.</i>Extensive evaluations on digital phantoms, physical phantoms, and clinical patient datasets demonstrate that the NNTV-GS method consistently outperforms conventional FDK and iterative SIRT-TV techniques across key metrics, including image quality, spatial resolution, and contrast-to-noise ratio. Ablation studies further validate the critical contribution of each component within the proposed framework. Notably, the method achieves high-quality 3D reconstructions within 10 min, offering a significant improvement in computational efficiency by reducing the optimization parameter space from hundreds of millions (voxel-based) to approximately one million (Gaussian-based), resulting in a two-orders-of-magnitude reduction in computational cost.<i>Significance.</i>This work represents a significant advancement for clinical dental CBCT imaging by enabling the use of iterative reconstruction algorithms that offer both high accuracy and faster reconstruction speeds. The reduction in reconstruction time while maintaining high precision meets the clinical demands for rapid and reliable imaging, thus contributing to enhanced patient care and improved diagnostic workflows.