Deep Learning Improves the Temporal Reproducibility of Aortic Measurement

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Surgery

Background:

• Imaging-based measurements are crucial for surgical decisions in aortic aneurysm patients.
• Manual measurements exhibit suboptimal temporal reproducibility, potentially leading to incorrect interventions.
• The need for improved reproducibility in thoracic aortic measurements is critical for patient management.

Purpose of the Study:

• To evaluate the hypothesis that deep learning can enhance the temporal reproducibility of CT angiography-derived thoracic aortic measurements.
• To compare the performance of deep learning segmentation against manual segmentation in terms of accuracy and reproducibility.
• To assess the clinical utility of deep learning for improving the reliability of aortic measurements.

Main Methods:

• A deep learning segmentation model was trained to extract aortic volume and diameter measurements from CT angiography.
• Non-inferiority of deep learning segmentation maps was confirmed by blinded cardiothoracic radiologists against manual segmentation.
• Temporal reproducibility was assessed using coefficient of reproducibility and standard deviation on longitudinal scan data (n=57 patients, 206 scans).

Main Results:

• Deep learning segmentation maps showed a slight preference over manual segmentation (p < 1e-5).
• Deep learning demonstrated superior temporal reproducibility for both aortic volume (p < 0.008) and diameter (p < 1e-5) measurements.
• Reproducibility metrics achieved by deep learning favorably compared with reported manual inter-rater variability.

Conclusions:

• Deep learning significantly improves the temporal reproducibility of thoracic aortic measurements derived from CT angiography.
• This AI-driven approach offers a more reliable method for assessing aortic aneurysm progression and guiding surgical decisions.
• The findings support future applications of deep learning in the quantitative evaluation of aortic diseases.