Intelligently Quantifying the Entire Irregular Dental Structure

Area of Science:

• Oral medicine
• Medical imaging
• Artificial intelligence

Background:

• Quantitative analysis of irregular oral anatomical structures is vital but challenging for clinicians.
• Current methods often rely on limited representative indicators, hindering comprehensive analysis.
• Deep learning semantic segmentation shows promise but faces hurdles with unclear boundaries and landmark acquisition.

Purpose of the Study:

• To develop an AI-powered measurement tool for the entire quantitative analysis of irregular dental structures.
• To address segmentation challenges and establish a clinical needs-compatible measurement coordinate system.
• To utilize lightweight networks for broader applicability and reduced computational demands.

Main Methods:

• Proposed an AI measurement tool utilizing the lightweight LU-Net model for semantic segmentation.
• Incorporated a compensation module to overcome segmentation difficulties caused by unclear boundaries.
• Performed additional enamel segmentation to establish a measurement coordinate system for quantitative analysis.

Main Results:

• Achieved high segmentation performance with Dice coefficients (0.934, 0.949) and IoU (0.888, 0.907) for palatal alveolar bone and enamel.
• Demonstrated no statistically significant difference from ground truth measurements, with low error metrics.
• Showed satisfactory agreement with manual measurements via Bland-Altman plots and intraclass correlation coefficients.

Conclusions:

• The novel intelligent approach enables swift and comprehensive quantitative analysis of irregular image structures.
• The tool facilitates personalized treatment planning, enhances clinical efficiency, and improves treatment success rates.
• This AI tool represents a significant advancement in the quantitative assessment of dental anatomy.