Fully automatic segmentation of type B aortic dissection from CTA images enabled by deep learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Surgery

Background:

• Type B aortic dissection (TBAD) requires precise anatomical measurements for effective treatment.
• Manual segmentation of TBAD from CT images is time-consuming and prone to inter-observer variability.

Purpose of the Study:

• To develop a fully automated and robust deep learning-based segmentation method for Type B aortic dissection (TBAD).
• To leverage convolutional neural network (CNN) models for accurate segmentation of the whole aorta, true lumen (TL), and false lumen (FL).

Main Methods:

• Retrospective collection of preoperative CT angiography (CTA) images from 276 TBAD patients.
• Establishment of a ground truth database using a manual segmentation protocol.
• Development and evaluation of three CNN models (single one-task, single multi-task, serial multi-task) using Dice coefficient score (DCS) and volume accuracy.

Main Results:

• The serial multi-task CNN (CNN3) demonstrated superior performance with high mean DCS values (0.91-0.93) for whole aorta, TL, and FL segmentation.
• CNN3 achieved excellent agreement between segmented and ground truth volumes, with minimal mean volume differences.
• The segmentation speed of CNN3 was significantly fast, averaging 0.038 seconds per image.

Conclusions:

• Deep learning offers a promising approach for accurate and efficient TBAD segmentation.
• The developed automated method enables precise measurements of TBAD anatomical features.
• This technology has the potential to improve the clinical management of TBAD patients.