Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Region of interest detection for efficient aortic segmentation

Loris Giordano, Ine Dirks, Tom Lenaerts, Jef Vandemeulebroucke

TL;DR

The paper tackles the challenge of efficient and accurate thoracic aorta segmentation on CT, where large field-of-view images and complex pathologies impose computational burdens on deep learning approaches. It introduces a compact ROI-detection module with a single anchor, integrated as a multi-task head at the bottleneck of an encoder–decoder segmentation network, and cascades detection with focused segmentation. On merged open datasets, the cascade achieves a mean DSC of $0.944$ with all cases above $0.9$, using roughly $1/3$ of the compute of full-image approaches. The method offers robust performance with reduced memory and inference time, enabling closer-to-clinical deployment and potential for on-site fine-tuning and quality-assurance workflows.

Abstract

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

Region of interest detection for efficient aortic segmentation

TL;DR

The paper tackles the challenge of efficient and accurate thoracic aorta segmentation on CT, where large field-of-view images and complex pathologies impose computational burdens on deep learning approaches. It introduces a compact ROI-detection module with a single anchor, integrated as a multi-task head at the bottleneck of an encoder–decoder segmentation network, and cascades detection with focused segmentation. On merged open datasets, the cascade achieves a mean DSC of with all cases above , using roughly of the compute of full-image approaches. The method offers robust performance with reduced memory and inference time, enabling closer-to-clinical deployment and potential for on-site fine-tuning and quality-assurance workflows.

Abstract

Thoracic aortic dissection and aneurysms are the most lethal diseases of the aorta. The major hindrance to treatment lies in the accurate analysis of the medical images. More particularly, aortic segmentation of the 3D image is often tedious and difficult. Deep-learning-based segmentation models are an ideal solution, but their inability to deliver usable outputs in difficult cases and their computational cost cause their clinical adoption to stay limited. This study presents an innovative approach for efficient aortic segmentation using targeted region of interest (ROI) detection. In contrast to classical detection models, we propose a simple and efficient detection model that can be widely applied to detect a single ROI. Our detection model is trained as a multi-task model, using an encoder-decoder architecture for segmentation and a fully connected network attached to the bottleneck for detection. We compare the performance of a one-step segmentation model applied to a complete image, nnU-Net and our cascade model composed of a detection and a segmentation step. We achieve a mean Dice similarity coefficient of 0.944 with over 0.9 for all cases using a third of the computing power. This simple solution achieves state-of-the-art performance while being compact and robust, making it an ideal solution for clinical applications.
Paper Structure (26 sections, 4 figures, 5 tables)

This paper contains 26 sections, 4 figures, 5 tables.

Table of Contents

  1. INTRODUCTION
  2. Clinical context
  3. Aortic segmentation models
  4. Context in large images
  5. Contributions
  6. MATERIALS AND METHODS
  7. Data
  8. Computational setup
  9. Detection model
  10. Ground-truth labels
  11. Detection model
  12. Complexity reduction
  13. Detection head
  14. Training and inference
  15. Segmentation model
  16. ...and 11 more sections

Figures (4)

  • Figure 1: Overview of our novel approach. We perform region of interest detection followed by focused segmentation of the cropped image to maximize output quality and minimize computational resources.
  • Figure 2: Region of interest extraction by applying a threshold on the normalized aortic mass along the axial direction. A non-linear transformation is applied to exaggerate the difference between 'heavy' and 'light' slices.
  • Figure 3: Two examples of the detection results. Red indicates the ground truth bounding box and segmentation. Blue indicates the predicted bounding box. The mean IoU is over 0.8, with complete containment of the thoracic aorta for each image.
  • Figure 4: Two examples of the segmentation results using the one-step, nnU-Net, and cascade approaches. The results are shown in the three main anatomical directions (from left to right: sagittal, coronal, axial) and as a 3D view of the segmentation. Red indicates the ground truth. Blue indicates the output segmentation. The mean Dice is the highest for the cascade approach, with values well over 0.9 and clear visual improvements.