Automatic Segmentation and Quantification of Abdominal Aortic Calcification in Lateral Lumbar Radiographs Based on...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Abdominal aortic calcification (AAC) is a significant indicator of cardiovascular risk.
• Accurate quantification of AAC is crucial for patient risk stratification.
• Current methods for AAC assessment can be time-consuming and subjective.

Purpose of the Study:

• To evaluate the performance of deep learning algorithms for automated segmentation and quantification of AAC.
• To compare the accuracy of a deep learning model against expert analysis and clinical reports.
• To assess the utility of U-Net models in analyzing lateral lumbar radiographs for AAC.

Main Methods:

• Retrospective analysis of 1359 lateral lumbar radiographs.
• Development of U-Net models for segmenting vertebrae T12-L5, aorta, and calcifications.
• Calculation of AAC lengths and Kauppila scores for automated quantification.
• Comparison of model predictions with an expert reference standard.

Main Results:

• The U-Net model achieved a high correlation (0.97) with the reference standard for total AAC score.
• Model accuracy for AAC severity was 0.77, compared to 0.74 for clinical reports.
• High concordance (Kendall's coefficient 0.89) was observed between model predictions and the reference standard.

Conclusions:

• Deep learning, specifically U-Net models, demonstrates high performance for automated AAC segmentation and quantification.
• The automated approach offers a reliable and potentially more efficient method for AAC assessment.
• This technology holds promise for improving cardiovascular risk assessment using routine radiographs.