Reliability of an Artificial Intelligence Software in the Detection of Approximal Caries Lesions Using Bitewing...

Area of Science:

• Dental Radiology
• Artificial Intelligence in Dentistry
• Caries Detection

Background:

• Evaluating the reliability of artificial intelligence (AI) software for detecting approximal caries lesions of varying depths on bitewing radiographs.
• Assessing AI's diagnostic capabilities against a consensus gold standard established by experienced dentists.

Purpose of the Study:

• To determine the accuracy and reliability of an AI software in identifying enamel and dentinal caries lesions from bitewing radiographs.
• To compare the AI software's detection results with a human expert consensus (gold standard).

Main Methods:

• 40 bitewing radiographs (288 teeth, 576 approximal surfaces) were analyzed.
• A consensus gold standard was established by five dentists using the International Caries Detection and Assessment System.
• AI software (Nostic software®) analyzed the radiographs, and its results were compared to the gold standard.

Main Results:

• The AI software achieved an accuracy of 0.78 for enamel lesions (D _1-2) and 0.85 for dentinal lesions (D _3-4).
• Area Under the Curve (AUC) values were 0.70 for enamel lesions and 0.81 for dentinal lesions.
• The AI demonstrated decent performance in detecting proximal caries lesions of different depths.

Conclusions:

• The AI software shows potential as an effective tool for supporting the diagnosis of initial caries in bitewing images.
• The AI's performance in detecting proximal caries lesions of varying depths is considered decent when compared to the gold standard.
• This technology could aid dental practitioners in early caries detection and diagnosis.