Self-Supervised Learning of Deep Embeddings for Classification and Identification of Dental Implants

Area of Science:

• Artificial Intelligence
• Biomedical Imaging
• Dental Implantology

Background:

• Automated detection of dental implants is crucial for accurate diagnosis and treatment planning.
• Existing methods often struggle with limited annotated data in dental radiography.
• Self-supervised learning, particularly masked image modeling (MIM), shows promise for feature learning without extensive labels.

Purpose of the Study:

• To develop an automated system for identifying dental implant systems using deep learning.
• To leverage and adapt masked image modeling (MIM) for self-supervised pre-training in dental imaging.
• To create a comprehensive dataset for dental implant design annotation.

Main Methods:

• Proposed a novel Masked Deep Embedding (MDE) pre-training method, an extension of the masked autoencoder (MAE) transformer.
• Employed self-supervised learning, specifically MIM, to overcome challenges with limited pre-training data in dentistry.
• Enhanced an existing dental dataset with expert annotations for implant components, creating the Implant Design Dataset (IDD).

Main Results:

• The MDE method achieved a superior average precision (AP) of 96.1 for dental implant detection.
• Outperformed supervised Vision Transformer (ViT) and standard MAE baselines by up to +2.9 AP.
• Successfully created the Implant Design Dataset (IDD) with detailed implant part annotations.

Conclusions:

• Self-supervised learning, particularly the MDE method, offers substantial performance improvements for dental implant detection with limited data.
• The developed system and dataset advance AI-driven solutions in implant dentistry.
• This work provides valuable tools for dentists and patients, enhancing implant-related diagnostics and care.