Objective: To investigate automatic assurance of image quality in cone beam CT (Cone-beam CT, CBCT) based on the Catphan®504 phantom. Methods: First, the CTP404 module was identified by the convolutional neural network. Then one slice is selected arbitrarily from the identified CTP404 modules and its inserts were located using the Circle Detection algorithm based on the Hough Transform. Finally, it was possible to measure automatically the CT number linearity, spatial resolution, low contrast resolution, contrast-to-noise ratio and uniformity index of the quality assurance parameter in the CBCT system using the number of slices obtained and the corresponding position of the inserts. Results: The accuracy of the classification results based on the convolutional neural network was 0.97, and the Hough Transfer algorithm was able to locate at least three inserts. Conclusion: Deep learning combined with traditional algorithms can automatically evaluate image quality in CBCT systems, which is of value for clinical applications.
Paper
Full text
Methods of Automatic Assurance of Image Quality in Cone-beam CT Based on Deep Learning
Semantic Scholar · Medicine · 2019
Abstract
Objective: To investigate automatic assurance of image quality in cone beam CT (Cone-beam CT, CBCT) based on the Catphan®504 phantom. Methods: First, the CTP404 module was identified by the convolutional neural network. Then one slice is selected arbitrarily from the identified CTP404 modules and its inserts were located using the Circle Detection algorithm based on the Hough Transform. Finally, it was possible to measure automatically the CT number linearity, spatial resolution, low contrast resolution, contrast-to-noise ratio and uniformity index of the quality assurance parameter in the CBCT system using the number of slices obtained and the corresponding position of the inserts. Results: The accuracy of the classification results based on the convolutional neural network was 0.97, and the Hough Transfer algorithm was able to locate at least three inserts. Conclusion: Deep learning combined with traditional algorithms can automatically evaluate image quality in CBCT systems, which is of value for clinical applications.