Evaluation of Color Changes in Composite Resins Using Artificial Intelligence-Based Digital Photography
TL;DR
Despite systematic bias in a* measurements, AI-assisted digital photography provided clinically acceptable color assessment under controlled conditions, supporting its potential as an adjunctive tool to spectrophotometric evaluation.
Abstract
This study aimed to compare color measurements of composite resins obtained using a spectrophotometer and a digital camera, and to evaluate the feasibility of an image-based classification model developed using Google Cloud Vertex Artificial Intelligence (AI) AutoML. Filtek Z350 XT (A2) specimens were immersed in a coffee solution at 37°C for 1, 3, or 5 hours, and CIE 1976 L*a*b* (CIELAB) color coordinates and ΔE*ab(stain) were recorded before and after immersion. Two AutoML models (VITA shade guide classification and ΔE*ab(stain) classification) were trained using standardized images. Digital photography yielded significantly lower a* values than spectrophotometry in all groups, whereas b* values did not differ significantly. Mean ΔE*ab(inter) values between instruments remained below 2.7 in all groups, indicating clinically acceptable agreement. Both AI models achieved an accuracy of 93.3%, exhibiting stable performance in identifying specimens with ΔE*ab(inter) < 2.7. Despite systematic bias in a* measurements, AI-assisted digital photography provided clinically acceptable color assessment under controlled conditions, supporting its potential as an adjunctive tool to spectrophotometric evaluation.