Deep Learning Based Airway Segmentation Using Key Point Prediction

The purpose of this study was to investigate the accuracy airway volume measurement by a Regression Neural Network-based deep-learning model. A set manually outlined data build algorithm for fully automatic segmentation deep learning process. Manual landmarks were determined one examiner using mid-sagittal plane cone-beam computed tomography (CBCT) images 315 patients. Clinical dataset-based training with augmentation conducted. Based on annotated landmarks, passage measured and segmented. our model confirmed measuring following between program: (1) difference in nasopharynx, oropharynx, hypopharynx, (2) Euclidean distance. For agreement analysis, 61 samples extracted compared. correlation test showed range good excellent reliability. volumes analyzed regression analysis. slope two measurements close 1 linear (r2 = 0.975, 1.02, p < 0.001). These results indicate that is possible via artificial intelligence. Additionally, high manual estimated.