Panoptic Segmentation on Panoramic Radiographs: Deep Learning-Based Segmentation of Various Structures Including...

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Panoramic radiographs (orthopantomograms) are essential dental diagnostic tools.
• Automated detection of diverse structures in these images is challenging due to variations in size and shape.
• Existing segmentation methods struggle with the complexity of structures in panoramic radiographs.

Purpose of the Study:

• To apply panoptic segmentation, integrating instance and semantic segmentation, to panoramic radiographs.
• To develop and evaluate a deep neural network for segmenting key anatomical structures and dental elements.
• To improve automated analysis of dental images for clinical applications.

Main Methods:

• Utilized a state-of-the-art deep neural network for panoptic segmentation.
• Trained the model to segment maxillary sinus, maxilla, mandible, mandibular canal, normal teeth, treated teeth, and dental implants.
• Implemented individual object classification for tooth and implant classes, distinguishing from semantic segmentation.
• Evaluated performance using panoptic quality, segmentation quality, recognition quality, IoU, and instance-level IoU.

Main Results:

• The deep learning-based AI model successfully performed panoptic segmentation on panoramic radiographs.
• Accurate segmentation was achieved for structures including the maxillary sinus and mandibular canal.
• Individual classification of teeth and implants demonstrated the model's ability to differentiate instances.

Conclusions:

• The developed AI model shows significant potential for automating the analysis of panoramic radiographs.
• Panoptic segmentation offers a robust approach for segmenting complex anatomical and dental structures.
• This automated method can assist dental practitioners in treatment planning and diagnosing maxillofacial diseases.