Feasibility of Artificial Intelligence-Processed Low-Dose Cone-Beam Computed Tomography in Dental Imaging

Key findings from this study

• The study found that AI-processed CBCT scans at 20% radiation dose achieved diagnostic quality equivalent to raw scans at full dose.
• The researchers demonstrate that AI processing of 10% dose images failed to achieve diagnostic acceptability despite enhancement attempts.
• The authors report that applying AI processing to full-dose images paradoxically reduced image quality compared to unprocessed scans.

Overview

This study evaluated whether artificial intelligence-based image processing can maintain diagnostic quality in low-dose dental cone-beam computed tomography scans. Radiation exposure from CBCT remains a clinical concern despite its diagnostic utility in dentistry. The research examined whether AI enhancement could offset image degradation from dose reduction across three dose levels.

Methods and approach

Researchers acquired CBCT scans from a single healthy adult male at 10%, 20%, and 100% of standard radiation dose. Each dataset underwent processing with an AI-based image enhancement model. Five dental specialists independently scored image quality using a 6-point scale across 12 criteria encompassing anatomical visibility, structural delineation, and diagnostic acceptability.

Results

AI-processed images at 20% dose showed no statistically significant difference from raw 100% dose images (median 4.45 vs. 5.05; p > 0.05). The 10% dose level produced significantly lower quality scores (p = 0.0074) despite AI processing. Raw 100% dose images received higher ratings than their AI-processed counterparts, indicating processing applied to full-dose scans may introduce artifacts or degradation.

Implications

These findings indicate moderate dose reduction coupled with AI enhancement may preserve diagnostic utility in CBCT imaging. The 20% dose threshold appears clinically viable when combined with appropriate image processing, potentially reducing cumulative radiation exposure without compromising diagnostic assessment. However, the inverse relationship between AI processing and image quality at standard dose levels warrants investigation into algorithm design and application parameters.

The preliminary nature of this work, based on a single subject, limits generalization to diverse patient anatomy and pathologies. Clinical implementation would require validation across larger, heterogeneous populations with varied tissue characteristics and dental conditions. Future studies must evaluate whether results hold across different CBCT systems and processing parameters.

Significant dose reduction to 10% remains unsustainable even with current AI processing. Further algorithm refinement targeting the 10-20% dose range may expand the clinical utility window. Investigation into why full-dose AI processing reduces image quality compared to raw images could inform model optimization strategies.

Scope and limitations

This summary is based on the study abstract and available metadata. It does not include a full analysis of the complete paper, supplementary materials, or underlying datasets unless explicitly stated. Findings should be interpreted in the context of the original publication.