Publication Details
Overview
 
 
 

Unpublished contribution to conference

Abstract 

Thoracic aortic dissection and aneurysms are the most lethal diseases of the aorta. The major hindrance totreatment lies in the accurate analysis of the medical images. More particularly, aortic segmentation of the 3Dimage is often tedious and difficult. Deep-learning-based segmentation models are an ideal solution, but theirinability to deliver usable outputs in difficult cases and their computational cost cause their clinical adoptionto stay limited. This study presents an innovative approach for efficient aortic segmentation using targetedregion of interest (ROI) detection. In contrast to classical detection models, we propose a simple and efficientdetection model that can be widely applied to detect a single ROI. Our detection model is trained as a multi-taskmodel, using an encoder-decoder architecture for segmentation and a fully connected network attached to thebottleneck for detection. We compare the performance of a one-step segmentation model applied to a completeimage, nnU-Net and our cascade model composed of a detection and a segmentation step. We achieve a meanDice similarity coefficient of 0.944 with over 0.9 for all cases using a third of the computing power. This simplesolution achieves state-of-the-art performance while being compact and robust, making it an ideal solution forclinical applications.

Reference