3D reconstruction from 2D multi-view dental 2D images based on EfficientNetB0 model.

Dental diseases are the primary cause of oral health concerns around the world, affecting millions of people. Therefore, recent developments in imaging technologies have transformed the detection and treatment of oral problems. Applying three-dimensional (3D) reconstruction from two-dimensional (2D) dental images, such as X-rays, is a potential development field. 3D reconstruction technology converts real-world goals into mathematical models that are compatible with computer logic expressions. It's been commonly used in dentistry. Particularly for patients with a vomiting reflex, 3D imaging techniques minimize patient discomfort and shorten the length of the examination or treatment. Therefore, this research paper proposes a new 3D reconstruction model from 2D multi-view dental images. The proposed framework consists of three stages. The first stage is the encoder stage, which extracts meaningful features from the 2D images. The second stage captures spatial and semantic information essential for the reconstruction task. The third stage is recurrence, which uses 3D long short-term memory (LSTM). It ensures that the information from various viewpoints is effectively integrated to produce a coherent representation of the 3D structure and decoder stage to translate the aggregated features from the LSTM into a fully reconstructed 3D model. When the proposed model was tested on the ShapeNet dataset, the suggested model achieved a maximum intersection over union (IoU) of 89.98% and an F1_score of 94.11%. A special case of 3D reconstruction, a dental dataset, has been created with the same structure as the ShapeNet dataset to evaluate our system. The proposed approach's results show promising accomplishments compared to many state-of-the-art approaches, and they motivate the authors to make plans for further improvement.