Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hierarchical Semantic Learning for Multi-Class Aorta Segmentation

Pengcheng Shi

TL;DR

This work tackles the challenge of multi-class aorta segmentation from CTA by introducing hierarchical semantic learning (HSL) integrated with curriculum learning (CL) and a fractal softmax framework. The Fractal Softmax operates on a tree-structured hierarchy to enforce semantic relationships across vessel classes, while a two-stage inference strategy and cbDice-based loss improve both speed and boundary fidelity. On the AortaSeg24 dataset, the method achieves substantial gains in validation Dice and surface metrics over baselines, with a reported $11.65\%$ Dice improvement on validation and a $5.6\%$ enhancement on the test set, alongside up to $5\times$ inference speedup. These advances enable more accurate and efficient real-time 3D vascular segmentation, facilitating planning for minimally invasive aortic repairs.

Abstract

The aorta, the body's largest artery, is prone to pathologies such as dissection, aneurysm, and atherosclerosis, which often require timely intervention. Minimally invasive repairs involving branch vessels necessitate detailed 3D anatomical analysis. Existing methods often overlook hierarchical anatomical relationships while struggling with severe class imbalance inherent in vascular structures. We address these challenges with a curriculum learning strategy that leverages a novel fractal softmax for hierarchical semantic learning. Inspired by human cognition, our approach progressively learns anatomical constraints by decomposing complex structures from simple to complex components. The curriculum learning framework naturally addresses class imbalance by first establishing robust feature representations for dominant classes before tackling rare but anatomically critical structures, significantly accelerating model convergence in multi-class scenarios. Our two-stage inference strategy achieves up to fivefold acceleration, enhancing clinical practicality. On the validation set at epoch 50, our hierarchical semantic loss improves the Dice score of nnU-Net ResEnc M by 11.65%. The proposed model demonstrates a 5.6% higher Dice score than baselines on the test set. Experimental results show significant improvements in segmentation accuracy and efficiency, making the framework suitable for real-time clinical applications. The implementation code for this challenge entry is publicly available at: https://github.com/PengchengShi1220/AortaSeg24. The code for fractal softmax will be available at https://github.com/PengchengShi1220/fractal-softmax.

Hierarchical Semantic Learning for Multi-Class Aorta Segmentation

TL;DR

This work tackles the challenge of multi-class aorta segmentation from CTA by introducing hierarchical semantic learning (HSL) integrated with curriculum learning (CL) and a fractal softmax framework. The Fractal Softmax operates on a tree-structured hierarchy to enforce semantic relationships across vessel classes, while a two-stage inference strategy and cbDice-based loss improve both speed and boundary fidelity. On the AortaSeg24 dataset, the method achieves substantial gains in validation Dice and surface metrics over baselines, with a reported Dice improvement on validation and a enhancement on the test set, alongside up to inference speedup. These advances enable more accurate and efficient real-time 3D vascular segmentation, facilitating planning for minimally invasive aortic repairs.

Abstract

The aorta, the body's largest artery, is prone to pathologies such as dissection, aneurysm, and atherosclerosis, which often require timely intervention. Minimally invasive repairs involving branch vessels necessitate detailed 3D anatomical analysis. Existing methods often overlook hierarchical anatomical relationships while struggling with severe class imbalance inherent in vascular structures. We address these challenges with a curriculum learning strategy that leverages a novel fractal softmax for hierarchical semantic learning. Inspired by human cognition, our approach progressively learns anatomical constraints by decomposing complex structures from simple to complex components. The curriculum learning framework naturally addresses class imbalance by first establishing robust feature representations for dominant classes before tackling rare but anatomically critical structures, significantly accelerating model convergence in multi-class scenarios. Our two-stage inference strategy achieves up to fivefold acceleration, enhancing clinical practicality. On the validation set at epoch 50, our hierarchical semantic loss improves the Dice score of nnU-Net ResEnc M by 11.65%. The proposed model demonstrates a 5.6% higher Dice score than baselines on the test set. Experimental results show significant improvements in segmentation accuracy and efficiency, making the framework suitable for real-time clinical applications. The implementation code for this challenge entry is publicly available at: https://github.com/PengchengShi1220/AortaSeg24. The code for fractal softmax will be available at https://github.com/PengchengShi1220/fractal-softmax.

Paper Structure

This paper contains 17 sections, 4 equations, 3 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Hierarchical Semantic Learning
  3. Curriculum Learning
  4. Methodology
  5. Fractal Softmax
  6. Two-Stage Inference
  7. Training Details
  8. Experiments and Results
  9. Dataset and Evaluation Protocol
  10. Implementation Details
  11. Results on validation set
  12. Results on Final Test Set
  13. Limitations and Future Work
  14. Conclusion
  15. Acknowledgments
  16. ...and 2 more sections

Figures (3)

  • Figure 1: The two-stage inference approach improves speed by up to 5 times compared to using only Model 2, depending on the expansion factor $m$ of the ROI in the first stage.
  • Figure 2: Hierarchical anatomical tree for AortaSeg24, illustrating superclass-subclass relationships from simple to complex structures.
  • Figure 3: Qualitative comparison of segmentation results. Left: Bad case showing discontinuities in small vessel categories (red arrows), likely due to limited training data for smaller structures. Right: Good case demonstrating accurate segmentation.