CT-based True- and False-Lumen Segmentation in Type B Aortic Dissection Using Machine Learning.

Area of Science:

• Medical imaging analysis
• Cardiovascular radiology
• Artificial intelligence in medicine

Background:

• Aortic dissection is a life-threatening condition requiring accurate imaging assessment.
• Quantitative analysis of true and false lumens is crucial for monitoring disease progression and guiding treatment.
• Current manual segmentation methods can be time-consuming and subjective.

Purpose of the Study:

• To develop and validate an automated segmentation pipeline for analyzing aortic dissection CT angiograms.
• To segment true and false lumens on multiplanar reformations (MPRs) perpendicular to the aortic centerline.
• To derive quantitative morphologic features, including aortic diameter and lumen cross-sectional areas.

Main Methods:

• Development of a two-convolutional neural network (CNN) based automated segmentation pipeline.
• Derivation of aortic centerline and generation of MPRs orthogonal to it.
• Training, validation, and testing of the CNN pipeline on 153 CT angiograms from 45 patients.

Main Results:

• The pipeline achieved high segmentation accuracy with mean Dice Similarity Coefficients (DSC) of 0.873 for true lumens and 0.894 for false lumens.
• Automated maximal diameter measurements showed excellent correlation with manual measurements (R² = 0.95).
• Derived cross-sectional area profiles of true and false lumens over time.

Conclusions:

• The developed segmentation pipeline accurately identifies true and false lumens in aortic dissection CT angiograms.
• This automated approach facilitates the quantitative assessment of aortic morphologic parameters.
• The derived parameters are valuable for patient surveillance and risk stratification in aortic dissection.