Unmet need

Panoramic X‑rays suffer from inherent ghost jaw, spinal overlay, and pharyngeal air‑gap artifacts that obscure teeth, mandibular canal, TMJ, and lesions, risking missed or uncertain diagnoses.​ Current enhancement tools, such as contrast equalization, super‑resolution, GAN-based deblurring improve image quality but do not explicitly correct these projection artifacts.​ Clinicians often require CBCT to clarify third molar–mandibular canal relationships and hidden pathologies, increasing cost, radiation dose, and workflow burden.​

The technology

• A two‑stage, fully automated AI pipeline
• Runs in approximately two seconds per radiograph
• Trained on expert‑annotated datasets​
• Leveraging an analogy between X‑ray attenuation and Retinex-based optical shadow models
• Yields artifact-corrected, high‑contrast panoramics without manual input.

Value proposition summary

• Substantially improves anatomical visibility: de‑shadowing enhances Weber contrast in the mandibular canal region by up to ~50% where ghost jaw artifacts overlap, revealing canal borders and lesion margins that were previously masked.​
• Demonstrates state‑of‑the‑art quantitative performance: artifact correction raises mandible and mandibular canal segmentation Dice scores (e.g., mandible Dice up to 0.9764 on a public dataset), outperforming multiple published methods even without enhancement.​
• Reduces need for additional imaging: clearer depiction of nerve–tooth relationships and lesion extent can obviate CBCT in selected cases, decreasing radiation exposure, cost, and time.​
•  Fully automated and scalable: high‑accuracy artifact segmentation (Dice up to ~0.96 for ghost jaw) yields downstream performance comparable to manual artifact labeling, eliminating labor‑intensive expert annotation at deployment.​
•  Flexible platform: the framework can incorporate alternative de‑shadowing backbones, adapt to new artifact types, and generalize across public and private datasets, supporting broad clinical and commercial translation

Figure 1. A panoramic radiograph (top) overlaid with artifact segmentations (bottom). From left to right: the ghost image of the opposite jaw, the spinal overlay, the pharyngeal air-gap.

Potential applications

• Clinical decision support in dentistry and oral & maxillofacial surgery – enabling clearer assessment of third molar proximity to the mandibular canal and TMJ morphology directly from panoramic X‑rays.​
• Pre‑processing module for AI diagnostic tools (e.g., tooth, mandible, and mandibular canal segmentation, lesion detection) to improve accuracy and robustness across devices and institutions.​
•  Integration into panoramic X‑ray systems or PACS as a real‑time image enhancement option for radiologists, surgeons, and general dentists.​
  
  Figure 2. The artifact removal process for panoramic X-rays. UNet++ is used to generate a mask of each artifact, which is then sent to a de-shadowing network. ShadowFormer enhances the image in artifact regions using the segmented masks. L1 differences are also shown.

Reference:
Dan et al (2025) Artifact Correction in Panoramic Radiographs using Deep De-shadowing,  ICCV Workshop