Exploratory analysis of Type B Aortic Dissection (TBAD) segmentation in 2D CTA images using various kernels

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Type-B Aortic Dissection (TBAD) is a rare, life-threatening condition.
• Accurate segmentation of aortic lumens in CT angiography (CTA) is crucial for diagnosis and treatment planning.
• Existing segmentation methods may require significant computational resources.

Purpose of the Study:

• To evaluate the feasibility of 2D Convolutional Neural Network (CNN) models for segmenting true lumen, false lumen, and false lumen thrombus in TBAD CTA images.
• To compare the performance of different 2D U-Net architectures against a 3D U-Net baseline and other segmentation models.
• To assess the potential of lightweight 2D models for real-time clinical decision-making.

Main Methods:

• Exploratory analysis of three 2D U-Net models: baseline, atrous convolution variant, and custom kernel variant.
• Training and benchmarking against a state-of-the-art 3D U-Net model.
• Performance evaluation using Dice and Intersection over Union (IoU) scores, compared with Segment Anything Model (SAM) and UniverSeg.

Main Results:

• The 2D U-Net with VGG19 encoder achieved the best performance among the tested 2D models (Dice: 80.48%, IoU: 72.93%).
• 2D U-Net models demonstrated high accuracy in segmenting true and false lumens.
• False lumen thrombus segmentation accuracy was lower compared to the 3D U-Net baseline.

Conclusions:

• 2D U-Net models, particularly with VGG19 encoder, show promise for accurate lumen segmentation in TBAD CTA.
• Challenges remain in achieving high accuracy for false lumen thrombus segmentation with 2D models.
• Development of lightweight 2D models is important for real-time applications and improved patient care in cardiovascular imaging.