Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Attention GhostUNet++: Enhanced Segmentation of Adipose Tissue and Liver in CT Images

Mansoor Hayat, Supavadee Aramvith, Subrata Bhattacharjee, Nouman Ahmad

TL;DR

The paper addresses automated segmentation of abdominal adipose tissue (SAT and VAT) and liver in CT images for body composition analysis. It introduces Attention GhostUNet++—a Ghost UNet++ architecture enhanced with Channel, Spatial, and Depth Attention in its bottlenecks—to refine features efficiently. On the AATTCT-IDS and LiTS datasets, it achieves Dice coefficients of approximately 0.9430 for VAT, 0.9639 for SAT, and 0.9652 for liver, outperforming several baselines while noting boundary segmentation challenges. The work promises scalable, annotation-friendly tools for large-scale BC studies and clinical workflows, with future directions including boundary refinement and cross-dataset generalization.

Abstract

Accurate segmentation of abdominal adipose tissue, including subcutaneous (SAT) and visceral adipose tissue (VAT), along with liver segmentation, is essential for understanding body composition and associated health risks such as type 2 diabetes and cardiovascular disease. This study proposes Attention GhostUNet++, a novel deep learning model incorporating Channel, Spatial, and Depth Attention mechanisms into the Ghost UNet++ bottleneck for automated, precise segmentation. Evaluated on the AATTCT-IDS and LiTS datasets, the model achieved Dice coefficients of 0.9430 for VAT, 0.9639 for SAT, and 0.9652 for liver segmentation, surpassing baseline models. Despite minor limitations in boundary detail segmentation, the proposed model significantly enhances feature refinement, contextual understanding, and computational efficiency, offering a robust solution for body composition analysis. The implementation of the proposed Attention GhostUNet++ model is available at:https://github.com/MansoorHayat777/Attention-GhostUNetPlusPlus.

Attention GhostUNet++: Enhanced Segmentation of Adipose Tissue and Liver in CT Images

TL;DR

The paper addresses automated segmentation of abdominal adipose tissue (SAT and VAT) and liver in CT images for body composition analysis. It introduces Attention GhostUNet++—a Ghost UNet++ architecture enhanced with Channel, Spatial, and Depth Attention in its bottlenecks—to refine features efficiently. On the AATTCT-IDS and LiTS datasets, it achieves Dice coefficients of approximately 0.9430 for VAT, 0.9639 for SAT, and 0.9652 for liver, outperforming several baselines while noting boundary segmentation challenges. The work promises scalable, annotation-friendly tools for large-scale BC studies and clinical workflows, with future directions including boundary refinement and cross-dataset generalization.

Abstract

Accurate segmentation of abdominal adipose tissue, including subcutaneous (SAT) and visceral adipose tissue (VAT), along with liver segmentation, is essential for understanding body composition and associated health risks such as type 2 diabetes and cardiovascular disease. This study proposes Attention GhostUNet++, a novel deep learning model incorporating Channel, Spatial, and Depth Attention mechanisms into the Ghost UNet++ bottleneck for automated, precise segmentation. Evaluated on the AATTCT-IDS and LiTS datasets, the model achieved Dice coefficients of 0.9430 for VAT, 0.9639 for SAT, and 0.9652 for liver segmentation, surpassing baseline models. Despite minor limitations in boundary detail segmentation, the proposed model significantly enhances feature refinement, contextual understanding, and computational efficiency, offering a robust solution for body composition analysis. The implementation of the proposed Attention GhostUNet++ model is available at:https://github.com/MansoorHayat777/Attention-GhostUNetPlusPlus.

Paper Structure

This paper contains 3 sections, 7 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Experimental Results
  3. Conclusion

Figures (2)

  • Figure 1: Attention GhostUNet++ Architecture.
  • Figure 2: Segmentation results for randomly selected CT image examples, the columns represent (1) the original CT image, (2) ground truth annotations, (3) model-predicted segmentation output, (4) mask differences between ground truth and predictions, (5) predicted segmentation masks overlaid on the original CT image.