Chest radiography deep radiomics-enabled aortic arch calcification interpretation across different populations

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Cardiovascular disease screening

Background:

• Early detection of aortic calcification is crucial for cardiovascular care planning.
• Opportunistic screening using chest radiography is a feasible approach for various populations.
• Aortic arch calcification is an indicator of cardiovascular risk.

Purpose of the Study:

• To develop and validate an artificial intelligence model for detecting aortic arch calcification on chest radiographs.
• To assess the model's performance in general and pre-end-stage kidney disease (pre-ESKD) populations.
• To identify regions indicative of aortic calcification in patients with and without pre-ESKD.

Main Methods:

• Utilized multiple deep convolutional neural network (CNN) transfer learning models.
• Fine-tuned pre-trained models and applied an ensemble technique for improved accuracy.
• Validated the approach on a derivation dataset and two external databases with distinct features.

Main Results:

• Achieved 84.12% precision, 84.70% recall, and an AUC of 0.85 in the general/older adult dataset.
• Obtained 87.5% precision, 85.56% recall, and an AUC of 0.86 in the pre-ESKD cohort.
• Identified specific regions on radiographs that discriminate between patients with and without pre-ESKD-related aortic calcification.

Conclusions:

• The developed ensemble AI model demonstrates high accuracy in detecting aortic arch calcification.
• The model shows potential for optimizing cardiovascular risk prediction when integrated into routine care.
• This approach can facilitate earlier detection and planning for cardiovascular care, particularly in at-risk populations like those with pre-ESKD.