Artificial Intelligence Models Accuracy for Odontogenic Keratocyst Detection From Panoramic View Radiographs: A...

Area of Science:

• Oral and Maxillofacial Radiology
• Artificial Intelligence in Medical Imaging
• Diagnostic Accuracy Studies

Background:

• Odontogenic keratocyst (OKC) is a common jaw lesion often misdiagnosed on panoramic radiographs.
• Accurate diagnosis is critical for effective patient management.
• Current manual interpretation of radiographs can lead to diagnostic inconsistencies.

Purpose of the Study:

• To systematically review and meta-analyze the diagnostic accuracy of AI models for detecting OKCs in panoramic radiographs.
• To compare the performance of different AI architectures in OKC detection.

Main Methods:

• Systematic literature search across 5 databases.
• Inclusion of studies evaluating AI models for OKC detection against reference standards.
• Extraction and pooling of sensitivity, specificity, and AUC using random-effects models.
• Meta-regression and subgroup analyses to explore heterogeneity; publication bias assessment.

Main Results:

• Eight studies were included in the meta-analysis.
• Pooled sensitivity was 83.66% and specificity was 82.89%.
• YOLO-based AI models achieved significantly higher performance (sensitivity 96.4%, specificity 96.0%).
• Model architecture was a significant factor in diagnostic performance, though high variability and publication bias were noted.

Conclusions:

• AI models, especially YOLO architectures, can enhance diagnostic accuracy for OKCs on panoramic radiographs.
• AI demonstrates strong potential for aiding in the diagnosis of simpler cases.
• AI should augment, not replace, human radiologist expertise, particularly in complex diagnostic scenarios.