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

Area of Science:

• Biomedical Imaging
• Computer Vision
• Dental Technology

Background:

• Dental diseases are a global health concern, necessitating advanced diagnostic tools.
• Two-dimensional (2D) dental imaging is common, but three-dimensional (3D) reconstruction offers enhanced visualization.
• Current 3D imaging methods can improve patient comfort and reduce examination times, especially for those with a vomiting reflex.

Purpose of the Study:

• To propose a novel 3D reconstruction model specifically for dental applications using 2D multi-view images.
• To enhance the accuracy and efficiency of dental diagnostics through advanced imaging techniques.
• To provide a more comfortable and less time-consuming experience for patients undergoing dental examinations.

Main Methods:

• A three-stage framework involving feature extraction (encoder), spatial-semantic information capture, and 3D reconstruction.
• Utilizing 3D long short-term memory (LSTM) for integrating multi-view information and generating a coherent 3D structure.
• Developing a dedicated dental dataset, structured similarly to ShapeNet, for model evaluation.

Main Results:

• The proposed model achieved a high intersection over union (IoU) of 89.98% on the ShapeNet dataset.
• An F1_score of 94.11% was attained, demonstrating the model's effectiveness in reconstruction tasks.
• Promising results were observed on the specialized dental dataset, outperforming several state-of-the-art approaches.

Conclusions:

• The developed 3D reconstruction model shows significant potential for advancing dental diagnostics.
• The approach offers a more patient-friendly alternative to traditional dental imaging methods.
• Further improvements and applications of this 3D reconstruction technique in dentistry are warranted.