Personalized prosthesis design in all-on-4® treatment through deep learning-accelerated structural optimization

Area of Science:

• Biomaterials Engineering
• Computational Mechanics
• Artificial Intelligence in Dentistry

Background:

• The All-on-4® treatment concept offers full-arch dental rehabilitation using four implants.
• The structural design of prosthetic frameworks for All-on-4® has received limited research attention.
• Optimizing framework design is crucial for material efficiency and structural strength.

Purpose of the Study:

• To propose and evaluate BESO-Net, a novel deep learning approach for optimizing denture framework structures.
• To reduce material consumption (e.g., Ti-6Al-4V) in All-on-4® prostheses while maintaining structural integrity.
• To accelerate the design process for personalized All-on-4® implant frameworks.

Main Methods:

• Development of BESO-Net, a bidirectional evolutionary structural optimization (BESO) based convolutional neural network (CNN).
• Training BESO-Net using finite element analysis (FEA) data from 14,994 design configurations.
• Evaluation of BESO-Net's performance, generalization, and computational efficiency.

Main Results:

• BESO-Net accurately predicted optimal denture framework structures across varied patient anatomies and load conditions.
• Achieved an average error of 0.29% for compliance and 11.26% for shape error compared to traditional methods.
• Reduced computational time for structural optimization from 6.5 hours to 45 seconds.

Conclusions:

• The BESO-Net approach is clinically applicable for rapid, personalized All-on-4® framework design.
• Significant material savings can be achieved without compromising framework stiffness.
• This AI-driven method enhances efficiency and material economy in dental implantology.