AI framework for dental shade matching under variable lighting conditions

Accurate tooth shade matching remains challenging because visual assessment is subjective and commercial spectrophotometers are sensitive to clinical conditions. Although artificial intelligence (AI) offers a promising alternative, most AI-based methods have been developed and tested under controlled imaging conditions, with limited evidence in routine practice. Herein, we developed PerceptShade, an AI framework for image-based shade estimation from routine intraoral photographs captured under variable lighting and with diverse devices, including smartphones and digital cameras.

PerceptShade incorporates clinical perceptual principles by focusing on non-specular tooth regions, enforcing a brightness-first hierarchical judgment inspired by standard clinical shade-matching protocols, and learning perceptual similarity through Perceptual Consistency Optimization. The framework was trained and evaluated using 1553 shade-matching images from 357 patients, with 50 prospectively collected independent cases used for an observer study. The dataset included heterogeneous lighting and device conditions to reflect routine clinical conditions.

PerceptShade achieved Top-1 and Top-3 accuracies of 79.12% and 96.44%, respectively, in the shade-matching test. In the independent observer dataset, PerceptShade achieved a Top-1 agreement of 76.0% with the Panel A consensus reference (95%CI: 61.8-86.9%), compared with 72.7% among independent expert observers (95%CI: 65.1%-80.3%). These Top-N recommendations may streamline digital shade-matching workflows and support clinical decision-making.

PerceptShade is a novel framework integrating hierarchical clinical reasoning with illumination-robust representation learning for dental shade matching. In this image-based evaluation, it demonstrated stable performance across heterogeneous acquisition conditions and may support shade selection in routine clinical photography.

PerceptShade is an AI-based dental shade-matching framework inspired by clinicians' perceptual workflow. Using routine photographs acquired without device calibration, it provides shade recommendations for single-tooth restorations. It may support candidate-shade selection, although it requires an in-frame reference tab and may still be affected by challenging imaging conditions.