Dentaldiff: Diffusion Probabilistic Models for Tumors and Cysts Segmentation in Dental Panoramic Radiographs.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate segmentation of dental abnormalities in panoramic radiographs is vital for diagnosis and treatment.
• Conventional Convolutional Neural Network (CNN)-based methods struggle with indistinct boundaries, complex structures, and imaging noise.

Purpose of the Study:

• To introduce Dentaldiff, a novel diffusion-based model for enhanced segmentation of tumors and cysts in dental panoramic radiographs.
• To improve segmentation performance and robustness against noise and complex anatomical variations.

Main Methods:

• Developed Dentaldiff, a diffusion-based segmentation model incorporating dynamic feature fusion and iterative denoising.
• Utilized a modified DenseUNet architecture to optimize segmentation accuracy.
• Evaluated performance using mean Intersection over Union (IoU) and Dice scores.

Main Results:

• Dentaldiff achieved state-of-the-art performance with a mean IoU of 0.61 ± 0.07 and Dice score of 0.75 ± 0.05.
• Outperformed existing CNN-based methods, especially in cases with indistinct boundaries and noise.
• Demonstrated effective handling of complex anatomical structures in dental radiographs.

Conclusions:

• Dentaldiff represents the first application of diffusion models to dental panoramic segmentation.
• The model shows significant potential for clinical application due to its superior performance in challenging segmentation tasks.
• This diffusion-based approach offers a promising advancement over traditional CNN methods for dental image analysis.