Fully automated image quality assessment based on deep learning for carotid computed tomography angiography: A...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Carotid computed tomography angiography (CTA) is crucial for diagnosing cerebrovascular diseases.
• Accurate image quality assessment (IQA) is vital for reliable CTA interpretation.
• Current IQA methods can be subjective and time-consuming.

Purpose of the Study:

• To develop and evaluate a fully automated deep learning and logistic regression model for IQA of carotid CTA images.
• To compare the model's performance against expert radiologist assessments.

Main Methods:

• Retrospective collection of 840 carotid CTA images.
• Radiologist assessment of image quality using a 3-point Likert scale.
• Development of an automated model using a 3D Res U-net and logistic regression on 600 training images.
• Validation on 240 internal and external test images.

Main Results:

• The automated multi-index model achieved excellent performance with AUCs of 0.98 (internal) and 0.97 (external).
• Consistency between the model and radiologists reached 91.8% and 92.6% in internal and external test datasets, respectively.
• The model demonstrated high sensitivity, specificity, precision, F-score, and accuracy.

Conclusions:

• The fully automated multi-index model offers performance equivalent to subjective radiologist evaluations for carotid CTA IQA.
• This automated approach enhances efficiency in image quality assessment.
• The model shows significant potential for routine clinical application in carotid CTA analysis.