What question did this study set out to answer?

This research aims to develop deep learning-based super-resolution models for cone-beam computed tomography images.

April 18, 2026Open Access

Deep learning-driven super-resolution for cone-beam computed tomography: An ex vivo proof-of-concept study using artificially degraded micro-computed tomography data

Key Points

This research aims to develop deep learning-based super-resolution models for cone-beam computed tomography images.
Artificially degraded micro-computed tomography data from extracted teeth used to simulate cone-beam CT images.
Three deep learning models (SRCNN, LTE, SwinIR) were compared with bicubic interpolation.
Image quality assessed using PSNR, SSIM, LPIPS, and DISTS metrics, alongside dentist evaluations using a Likert scale.
All deep learning models significantly surpassed bicubic interpolation on objective quality metrics.
SwinIR achieved the highest PSNR, while SRCNN obtained the best SSIM.
LTE excelled in LPIPS and DISTS metrics, and received the highest sharpness ratings from dentists.
LTE showed improved crack visibility in images, with significant statistical support.

Abstract

Purpose: This ex vivo proof-of-concept study aimed to develop deep learning (DL)-based super-resolution (SR) models to enhance simulated cone-beam computed tomography (CBCT) images.Materials and Methods: Micro-computed tomography data from 51 extracted teeth were artificially degraded to simulate CBCT images.Three DL models, super-resolution convolutional neural network (SRCNN), local texture estimator (LTE), and Swin Transformer for image restoration (SwinIR), were compared with bicubic interpolation.Image quality was assessed using peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), learned perceptual image patch similarity (LPIPS), and deep image structure and texture similarity (DISTS).Three dentists evaluated sharpness and noise using a 5-point Likert scale.Eight observers assessed crack visibility in 47 images for LTE and bicubic interpolation using a 5-point Likert scale; scores were binarized using high and low thresholds.Results: All models significantly outperformed bicubic interpolation on objective metrics.SwinIR showed the highest PSNR (30.362.66),whereas SRCNN achieved the highest SSIM (0.8890.073).LTE achieved the best LPIPS (0.2530.101) and DISTS (0.2030.049).Subjectively, LTE received the highest sharpness ratings (mean.3.790.47),whereas bicubic interpolation received the highest noise ratings (3.971.43).LTE significantly improved crack visibility (odds ratio = 1.326,P = 0.006 for the low-threshold analysis; odds ratio = 1.310,P = 0.010 for the highthreshold analysis), with a higher pooled area under the curve (0.81 vs. 0.76, P = 0.063).Conclusion: DL-based SR models can enhance simulated CBCT images, with LTE demonstrating superior perceptual sharpness and crack visibility.

Bookmark

View Full Paper

Cite This Study

Mohammad‐Rahimi et al. (Thu,) studied this question.

synapsesocial.com/papers/69e31ec840886becb653e7ec https://doi.org/https://doi.org/10.5624/isd.20260010

Bookmark

View Full Paper