Precise sparse-view dental crown reconstruction via 3D Gaussian neural fields

Accurate three-dimensional reconstruction of dental crowns is crucial for improving the effectiveness of orthodontic diagnosis and treatment. However, in the face of a patient population that makes it difficult to undergo routine face-to-face consultations, utilizing extremely limited intraoral images to achieve high-precision crown model reconstruction to meet the urgent need for remote dental monitoring remains a major challenge in current clinical practice and computer vision fields. To overcome the above challenges, this article innovatively proposes the DGNF-DC method, a sparse view dental crown accurate reconstruction framework based on a three-dimensional Gaussian neural field. Its core contribution lies in: (1) Target segmentation and image enhancement preprocessing, optimizing input data quality; (2) Building a robust 3D Gaussian representation of dental crown foundation using a ControlNet++model that integrates camera pose estimation and optimization; (3) Introducing learnable neural representations to drive 3D Gaussian element densification of dental crown foundations and distill geometric details; (4) Based on learnable neural representation for geometric perception sampling, combined with multi-scale Gaussian rendering strategy, high fidelity dental crown model reconstruction is achieved. Numerous experimental validations have shown that the DGNF-DC method is significantly superior to the current mainstream methods and exhibits excellent and robust reconstruction performance under different input image quantities. Significant improvements have been achieved in key evaluation indicators: PSNR increased by 1.32 %, SSIM increased by 3.6 %, and LPIPS decreased by 6.7 %. DGNF-DC provides practical and reliable technical support for remote dental monitoring and precise orthodontic diagnosis and treatment, effectively promoting the practical process of digital dental diagnosis and treatment.