Related Experiment Video
Updated: Aug 6, 2026
Super-Resolution Imaging and Shared Management: A Protocol for Confocal Microscopy with Multiplex Detection
Published on: February 24, 2026
Validating the Utility of Super-Resolution for Downstream Tasks in Periapical Films
Junran Peng1, Meiyu Hu2, Qianli Zhang3
1School of Computer and Communication Engineering, University of Science and Technology Beijing, Beijing, China; Shunde Innovation School, University of Science and Technology Beijing, Foshan, China.
Introduction And Aims:
Periapical radiography is widely used in dental practice, but its diagnostic yield is often compromised by motion artefacts, sensor limitations, and storage compression. This study evaluates whether a deep-learning-based super-resolution (SR) framework can enhance degraded periapical films to improve automated diagnostic accuracy without additional radiation exposure.
Methods:
A clinical evaluation pipeline was established using 6283 periapical films. First, 5998 images were used to train an artificial intelligence-driven restoration model. Then, 285 meticulously annotated clinical images validated the SR preprocessing impact. We compared automated diagnostic performance on original low-resolution images vs SR-enhanced reconstructions, focusing on multistructure semantic segmentation and pathology/restoration detection.
Results:
SR-enhanced images consistently outperformed baseline inputs. For segmentation, SR increased mean intersection over union (IoU) from 55.17% to 60.24% (P < .0001) and boundary precision (boundary IoU) improved from 23.64% to 48.22% (P < .0001), with significant improvements observed across tooth structure, restorative, and pathological category groups, demonstrating superior margin delineation. In detection tasks, SR increased recall by 4.24% (P = .0121), while mean Average Precision (mAP@50) showed a modest nonsignificant increase from 67.18% to 69.30% (P = .1114), indicating that SR primarily reduces missed detections rather than uniformly elevating detection precision.
Conclusions:
Deep-learning-driven SR effectively recovers critical diagnostic details from suboptimal periapical films. By significantly enhancing boundary sharpness and reducing missed detections, SR serves as an effective preprocessing step that improves geometric precision and reliability of automated dental analyses.
Clinical Relevance:
This software-based enhancement standardizes image quality and elevates diagnostic accuracy cost-effectively, particularly in resource-constrained settings. It maximizes the diagnostic utility of routine 2D periapical radiographs, supporting precise clinical decision-making without the radiation burden of 3D imaging.
