Automated Detection and Segmentation of Ascending Aorta Dilation on a Non-ECG-Gated Chest CT Using Deep Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Cardiovascular Imaging

Background:

• Ascending aortic dilation is a key risk factor for aortic dissection.
• Dilation often goes undetected in routine, non-ECG-gated chest CT scans.
• Automated detection methods are needed for opportunistic screening.

Purpose of the Study:

• Develop and evaluate a deep learning pipeline for automated ascending aortic segmentation.
• Utilize non-ECG-gated chest CT scans for analysis.
• Improve detection rates of ascending aortic dilation.

Main Methods:

• A two-stage deep learning pipeline was designed.
• Incorporated a convolutional neural network (CNN) for slice classification.
• Employed a U-Net model for aortic segmentation on 500 non-ECG-gated CT scans.

Main Results:

• Achieved high performance on a test set: Dice Similarity Coefficient (DSC) of 99.21% and Intersection over Union (IoU) of 98.45%.
• Focus-slice classification accuracy reached 98.18%.
• Outperformed traditional and prior CNN-based methods with higher overlap metrics and low computational cost.

Conclusions:

• A lightweight CNN+U-Net model enables accurate, automated ascending aortic segmentation.
• The pipeline can enhance diagnostic accuracy for aortic dissection risk.
• Facilitates opportunistic detection of ascending aortic dilation in routine CT imaging, reducing radiologist workload.