Human Tooth Crack Image Analysis with Multiple Deep Learning Approaches.

Area of Science:

• Dental Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Tooth cracks are common and difficult to detect, often requiring prompt treatment to prevent tooth loss.
• Traditional imaging methods like X-rays have limitations in visualizing subtle tooth cracks.
• Indocyanine green (ICG) assisted near-infrared fluorescence (NIRF) imaging offers improved visualization due to deep light penetration and ICG fluorescence.

Purpose of the Study:

• To evaluate the efficacy of an ICG-assisted NIRF dental imaging technique for detecting tooth cracks.
• To apply machine learning and deep learning models for automated analysis of tooth crack images.
• To assess the potential of this combined approach in assisting dentists with crack diagnosis.

Main Methods:

• A dataset of 593 cracked and 601 non-cracked tooth images was acquired using NIR imaging videos.
• Machine learning models, including pre-trained residual networks and squeezenet1_1, were used for image classification.
• Object detection (Single Shot Multi-Box Detector - SSD) and super-resolution (SR-Generative Adversarial Network) models were employed for crack recognition and image enhancement.

Main Results:

• Classification models achieved high accuracy: 88.2% for residual network and 94.25% for squeezenet1_1 in distinguishing cracked from non-cracked teeth.
• The SSD model successfully recognized cracks regardless of input image size.
• The SR-Generative Adversarial Network demonstrated improved resolution of crack images.

Conclusions:

• Deep learning models significantly enhance the efficiency and accuracy of tooth crack identification.
• The developed NIR dental imaging system, coupled with deep learning, shows strong potential to assist dentists in diagnosing tooth cracks.
• This technology could lead to earlier and more reliable detection of dental fractures.