Deep Learning-Assisted Localization of Cystic Lesions and Benign Tumors in the Maxillofacial Region Using Panoramic...

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Automated detection of cystic lesions and benign tumors in the maxillofacial region using panoramic radiographs can aid early recognition.
• This study focuses on the feasibility of deep learning for localizing specific jaw pathologies.

Purpose of the Study:

• To develop and evaluate a Mask R-CNN deep learning model for localizing dentigerous cysts (DCs), radicular cysts (RCs), odontogenic keratocysts (OKCs), and ameloblastomas on panoramic radiographs.

Main Methods:

• A Mask R-CNN model was trained on 184 pathology-confirmed lesions from 215 panoramic radiographs.
• The model was tested on 47 lesions, evaluating performance using precision, sensitivity, and F1 score at an IoU threshold of 0.1.

Main Results:

• The model achieved an overall sensitivity of 55.3% and a precision of 83.9% on the test set (n=47).
• Lesion-specific sensitivities varied: 93.3% for DCs, 40.0% for ameloblastomas, 37.5% for OKCs, and 36.8% for RCs.

Conclusions:

• A deep learning approach shows preliminary feasibility for localizing jaw lesions on panoramic radiographs.
• Limitations include a lack of control images and a small dataset, restricting generalizability. Larger, balanced datasets are necessary for clinical validation.