Overjet Caries Assist

Overjet Caries Assist is a software tool employing machine learning algorithms to analyze bitewing and periapical dental radiographs for enhanced caries detection and segmentation. It aims to serve as an adjunctive diagnostic aid for dentists, with studies demonstrating its ability to improve their sensitivity in identifying carious lesions.

Overjet Caries Assist (OCA) is a radiological, automated, computer-assisted detection (CADe) software designed to aid dentists in the detection and segmentation of caries on bitewing and periapical radiographs. It functions as an adjunctive tool, providing additional information for diagnosis but not replacing the dentist's comprehensive review and clinical judgment. The software operates using a three-layer architecture (Network, Decision, Presentation). The Decision Layer employs a machine learning system with four modules: an Image Preprocessor (resizes, normalizes, classifies image type), a Tooth Number Assignment Module (identifies teeth and provides segmentation masks), a Caries Module (segments carious lesions using an ensemble of three Instance Segmentation models), and a Post Processing module that annotates the original radiograph with predicted carious lesions assigned to specific teeth. This version of OCA expands its utility to periapical radiographs, more sensor manufacturers, and patients aged 12 and older with permanent dentition. Performance testing demonstrated OCA's efficacy. Standalone algorithm performance, evaluated against a ground truth established by dental professionals, showed overall sensitivity of 76.6% for bitewing and 79.4% for periapical images, with specificities of 99.1% and 99.4% respectively. Mean Dice scores for lesion segmentation were generally above 0.73 for various caries types and image modalities. A clinical reader study involving 28 dentists revealed that OCA assistance significantly improved their caries detection sensitivity: from 64.6% to 78.5% for bitewing images, and from 65.6% to 79.0% for periapical images. While a slight decrease in specificity (0.4% for both image types) was observed, the weighted Alternative Free Response Receiver Operating Characteristic (wAFROC) scores indicated a statistically significant overall improvement in detection performance (AUC increase of 0.055 for bitewing, 0.050 for periapical) when dentists used OCA, highlighting its benefit.

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 ensemble of three Instance Segmentation models in the Caries Module specifically contribute to improved robustness or accuracy in caries detection compared to a single, potentially more complex, model?; What were the specific criteria or thresholds used by the oral radiologist for adjudicating non-consensus labels in establishing the ground truth, and how might variations in these criteria affect the reported performance metrics?; The document states the device is trained not to predict caries in areas of significant tooth overlap; what is the quantitative definition of 'significant overlap,' and what percentage of tooth surfaces in the test datasets were excluded from analysis due to this criterion?; Could the slight decrease in specificity observed in the reader study (0.4%) translate to a clinically relevant increase in false positives, and were the types or characteristics of these false positives analyzed?; What is the performance of the Tooth Number Assignment Module, particularly in cases of missing teeth, supernumerary teeth, or unusual dental anatomy, and how do errors in this module propagate to the Caries Module?