Deep learning for dentomaxillofacial cone-beam computed tomography image quality enhancement: A pilot study

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Cone-beam computed tomography (CBCT) is crucial in dentomaxillofacial imaging.
• Image quality limitations, such as noise and low resolution, can impact diagnostic accuracy.
• Enhancing CBCT image quality is essential for improved clinical outcomes.

Purpose of the Study:

• To develop and evaluate a deep learning-based super-resolution method for CBCT images.
• To improve the quality of dentomaxillofacial CBCT images using artificial intelligence.
• To assess the effectiveness of the MIRNet-v2 model in enhancing CBCT image resolution and reducing noise.

Main Methods:

• A deep learning super-resolution model (MIRNet-v2) was developed using 6,961 CBCT slices.
• Low-resolution images were created via downscaling, blurring, and noise addition.
• Model performance was evaluated using peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and radiologist assessments.

Main Results:

• The deep learning model significantly improved CBCT image quality across all metrics.
• Enhanced images achieved mean PSNR > 35 dB and SSIM > 0.85, particularly for blurred images (PSNR: 43.86, SSIM: 0.98).
• Radiologists noted improvements in diagnostic quality, noise reduction, and spatial resolution, with comparable outputs to original images.

Conclusions:

• Deep learning super-resolution holds significant potential for enhancing CBCT image quality, especially in blurred or downscaled scenarios.
• This technology may enable lower radiation doses in CBCT imaging.
• Improved CBCT image quality can lead to more accurate diagnoses and better clinical decision-making.