Videa Caries Assist

Videa Caries Assist is an AI-powered software tool that employs supervised deep learning to analyze dental bitewing radiographs for the detection and localization of carious lesions in adult patients. Clinical studies demonstrate that its use as a computer-assisted detection aid significantly improves the diagnostic accuracy of dentists by enhancing lesion identification compared to unaided radiographic review.

Videa Caries Assist is a cloud-based, AI-powered software functioning as a computer-assisted detection (CADe) device for dental professionals. It utilizes supervised deep learning algorithms to automatically analyze intraoral bitewing radiographs from adult patients (22 years and older), identifying and localizing all types of carious lesions across various tooth surfaces. Accessed via an API through a dental viewer, the system provides a binary indication of lesion presence and, if found, outputs bounding box coordinates for each detected lesion, serving as a concurrent review tool for licensed dentists. Performance evaluation involved standalone testing and a clinical reader study. The standalone assessment on 1034 radiographs achieved an overall average Alternative Free-response Receiver Operating Characteristic Figure of Merit (AFROC FOM) of 0.740, with 70.8% image-based sensitivity and 59.5% image-based positive predictive value. The multi-reader clinical study (226 radiographs, 21 readers) demonstrated superior diagnostic accuracy when dentists used Videa Caries Assist, with the mean AFROC FOM increasing from 0.667 (unaided) to 0.739 (aided), a statistically significant improvement. This enhancement included increased sensitivity and a reduction in the non-lesion fraction, though a slight, non-statistically significant rise in the false positive fraction was observed with device assistance.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the 'supervised deep learning' algorithm of Videa Caries Assist differentiate between various types and severities of carious lesions (e.g., incipient vs. advanced, active vs. arrested), and what specific features from the radiographs does it primarily rely on for detection?; The reader study showed a statistically significant improvement in AFROC FOM with Videa Caries Assist; what was the impact on inter-reader variability, and did the device disproportionately benefit less experienced dentists more than highly experienced ones?; Given the standalone AFROC FOM of 0.740 and the aided reader AFROC FOM of 0.739, what factors contribute to the AI's standalone performance being similar to, or even slightly higher than, the AI-assisted human performance in this specific metric?; The document mentions a slightly lower AFROC FOM for the Schick 33 sensor in standalone testing. Although reader study sub-analysis alleviated concerns, what are the specific image characteristics of Schick 33 radiographs that might pose a challenge for the AI, and are there ongoing efforts to optimize performance across all sensor types?; While the False Positive Fraction (FPF) increase with aided reads was not statistically significant, what is the clinical impact of any increase in false positives on workflow efficiency and potential for unnecessary further investigation, and how does this compare to the benefit of increased sensitivity?