Evaluation of artificial intelligence for airway analysis on cone beam computed tomography

Aim

The aim of the study is to verify the success of an artificial intelligence model for the automatic airway segmentation on cone beam computed tomography (CBCT) images.

Materials and Methods

Three hundred CBCT images of adults were used in this study for airway assessment. The algorithm development was carried out using the Mask R-CNN ResNet 101 model. Both manual segmentation and an artificial intelligence (AI) system from Velmeni, Inc., were used for airway analysis. The airway analysis was determined by two oral and maxillofacial radiologists using Anatomage InVivo 3D software. Additionally, the convolutional neural network−based architecture was employed for airway volume detection. A comparison was made between the results obtained from the human observers and the artificial intelligence model.

Results

In evaluating the performance of the AI model for the segmentation of airway analysis, true positive, false positive, and false negative values were found to be 485, 18, and 23, respectively. Sensitivity, precision, and F1 score values were calculated as 0.9332, 0.9615, and 0.9766, respectively. The area under curve value was calculated as 0.8467.

Conclusion

The integration of the AI Mask R-CNN ResNet 101 model for airway analysis holds great promise in the decision support system for diagnostic accuracy, treatment planning, and overall treatment outcomes.