Detection of maxillary sinusitis of endodontic origin in cone-beam CT images using deep learning algorithms.
Authors
Affiliations (7)
Affiliations (7)
- The British University in Egypt (BUE), El Sherouk City, Egypt.
- Cairo University, Cairo, Egypt.
- Egyptian Maxillofacial Radiology Alliance Lab, Cairo, Egypt.
- The British University in Egypt (BUE), El Sherouk, Cairo, Egypt.
- University of Pennsylvania, Philadelphia, PA, United States of America.
- The British University in Egypt (BUE), El Sherouk, Cairo, Egypt. [email protected].
- The British University in Egypt (BUE), El Sherouk City, Egypt. [email protected].
Abstract
Cone-Beam Computed Tomography (CBCT) scans were retrospectively collected and examined to acquire a balanced dataset representative of Normal Maxillary Sinus (NMS), Maxillary Sinusitis of Endodontic origin (MSEO), and Maxillary Sinusitis of Non-Endodontic origin (MS-NEO) according to established criteria. Data were manually labeled, pre-processed and split into training, validation, and testing groups. A custom model workflow started by anatomic classification of input images as being coronal, sagittal, or axial, followed by segmentation of roots, lining, and sinus regions, followed by extraction of radiographic sinus features for final classification of NMS, MSEO, or MS-NEO. Model performance on testing and external datasets was performed evaluating Accuracy, Precision, Recall, F1, and DICE scores. Results showed that the anatomic classifier achieved overall Accuracy, Precision, Recall, and F1 scores of 0.99 and 0.98 on the testing and external datasets. The overall DICE scores for sagittal, coronal, and axial segmenters were 0.8, 1, and 0.9 on the testing, and 0.8, 0.9, and 0.9 on the external dataset. The features extractor achieved overall Accuracy and F1 scores of 0.9-1 for sagittal, coronal, and axial classifiers for testing and external datasets, while the Multi-View classifier exhibited excellent metrics for both datasets with potential to improve clinical diagnosis of MSEO.