Deep learning-based automated segmentation and quantification of aortic arch calcification at chest radiograph

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease

Background:

• Aortic arch calcification (AoAC) is a significant predictor of coronary heart disease and cardiovascular outcomes.
• Automated detection and quantification of AoAC from chest radiographs (X-rays) are crucial for risk assessment.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network for automated AoAC detection and quantification from routine chest radiographs.
• To assess the association between AoAC and clinical variables, including hypertension.

Main Methods:

• A deep-learning segmentation model was trained on radiologist-annotated chest radiographs for AoAC contour acquisition.
• The model's performance was evaluated using AUC, accuracy, sensitivity, and specificity.
• Radiomics features from segmented AoAC regions were analyzed, and multivariable logistic regression assessed AoAC associations with clinical factors.

Main Results:

• The study included 347 patients (median age 77 years, 224 women).
• Radiomics parameters of the aorta arch correlated with AoAC severity.
• AoAC was independently associated with hypertension (OR = 26.48, P = 0.002).

Conclusions:

• An automated framework for AoAC segmentation and quantification on chest radiographs was developed.
• This tool can be utilized in community checkups for rapid AoAC assessment and cardiovascular risk evaluation.