[Paper Review] Directional Connectivity-based Segmentation of Medical Images

Ziyun Yang, Sina Farsiu|arXiv (Cornell University)|Mar 31, 2023

Radiomics and Machine Learning in Medical Imaging9 citations

TL;DR

The paper introduces a directional connectivity modeling scheme that disentangles directional sub-space in the latent space to improve anatomically consistent biomarker segmentation, showing improvements over state-of-the-art on public benchmarks.

ABSTRACT

Anatomical consistency in biomarker segmentation is crucial for many medical image analysis tasks. A promising paradigm for achieving anatomically consistent segmentation via deep networks is incorporating pixel connectivity, a basic concept in digital topology, to model inter-pixel relationships. However, previous works on connectivity modeling have ignored the rich channel-wise directional information in the latent space. In this work, we demonstrate that effective disentanglement of directional sub-space from the shared latent space can significantly enhance the feature representation in the connectivity-based network. To this end, we propose a directional connectivity modeling scheme for segmentation that decouples, tracks, and utilizes the directional information across the network. Experiments on various public medical image segmentation benchmarks show the effectiveness of our model as compared to the state-of-the-art methods. Code is available at https://github.com/Zyun-Y/DconnNet.

Motivation & Objective

• Promote anatomically consistent biomarker segmentation by leveraging pixel connectivity.
• Incorporate and disentangle channel-wise directional information in latent space for connectivity-based networks.
• Develop and evaluate a directional connectivity modeling scheme across medical image segmentation tasks.

Proposed method

• Decouple, track, and utilize directional information across the segmentation network.
• Model connectivity by integrating directional sub-space information with the shared latent space.
• Leverage channel-wise directional cues in latent representations to improve connectivity-based segmentation.

Experimental results

Research questions

• RQ1 Does incorporating directional connectivity improve segmentation accuracy and anatomical consistency?
• RQ2 Can disentangling directional sub-space in the latent space enhance feature representation for connectivity-based segmentation?
• RQ3 How does the proposed directional connectivity method perform on public medical image segmentation benchmarks compared to state-of-the-art methods?

Key findings

• The proposed directional connectivity model demonstrates effectiveness on public medical image segmentation benchmarks.
• Shows performance gains over state-of-the-art methods on evaluated datasets.
• Provides evidence that disentangling directional information in latent space benefits connectivity-based segmentation.