Current 3D Gaussian Splatting (GS) reconstruction methods in endoscopy struggle with high similarity among endoscopy images and large dynamic changes caused by intestinal peristalsis. To address these limitations, we introduce the Dual-Domain Deformation Model ( \(\hbox {D}^{3}\hbox {M}\) ), a novel approach that accurately models the intestinal lining with dynamic and complex deformations. \(\hbox {D}^{3}\hbox {M}\) innovates by operating in both spatial and temporal domains, enabling robust tracking and reconstruction of the intestinal surface despite significant motion and appearance changes. Unlike existing methods, \(\hbox {D}^{3}\hbox {M}\) employs a unique dual-domain framework that enhances its ability to capture intricate deformations and maintain reconstruction accuracy. Through extensive experiments on the EndoNeRF and StereoMIS dataset, we demonstrate that \(\hbox {D}^{3}\hbox {M}\) significantly outperforms existing 3D-GS techniques, achieving higher reconstruction accuracy and robustness. Quantitative metrics reveal substantial improvements in error rates and computational efficiency, validating our claim of superior performance. These results highlight \(\hbox {D}^{3}\hbox {M}\) ’s effectiveness in capturing the complex dynamics of the intestinal lining, offering a more accurate and reliable 3D reconstruction. This advancement has the potential to revolutionize endoscopic imaging by providing detailed and dynamic visualizations of the intestinal environment, thereby enhancing clinical utility. Future work will focus on clinical validation to assess the practical impact of \(\hbox {D}^{3}\hbox {M}\) in real-world endoscopic procedures. Additionally, we will explore the scalability and adaptability of \(\hbox {D}^{3}\hbox {M}\) to handle different types of endoscopic procedures and other medical imaging modalities, ensuring its broad applicability in diverse clinical settings.
