CT-based brain ventricle segmentation via diffusion Schrödinger Bridge without target domain ground truths

TL;DR

The paper tackles CT-based brain ventricle segmentation under scarce labeled CT data by leveraging diffusion-model-based domain adaptation to translate MRI ventricle labels to CT, enabling joint learning of image translation and segmentation without target-domain ground truths. An Unpaired Neural Schrödinger Bridge (UNSB) guides MRI-to-CT translation, paired with an Attention Recurrent Residual U-Net (R2AUNet) for segmentation and Monte Carlo dropout to produce uncertainty maps. Compared against CycleGAN, CUT, SynSeg-Net, and two-stage baselines, the proposed end-to-end framework achieves a Dice score of $0.78 \pm 0.27$ and outperforms baselines, while uncertainty estimates facilitate quality control. The approach holds promise for improved ventriculostomy planning in emergent settings and could extend to other CT-based neuroanatomical segmentation tasks, with future work expanding to 3D processing and broader clinical data.

Abstract

Efficient and accurate brain ventricle segmentation from clinical CT scans is critical for emergency surgeries like ventriculostomy. With the challenges in poor soft tissue contrast and a scarcity of well-annotated databases for clinical brain CTs, we introduce a novel uncertainty-aware ventricle segmentation technique without the need of CT segmentation ground truths by leveraging diffusion-model-based domain adaptation. Specifically, our method employs the diffusion Schrödinger Bridge and an attention recurrent residual U-Net to capitalize on unpaired CT and MRI scans to derive automatic CT segmentation from those of the MRIs, which are more accessible. Importantly, we propose an end-to-end, joint training framework of image translation and segmentation tasks, and demonstrate its benefit over training individual tasks separately. By comparing the proposed method against similar setups using two different GAN models for domain adaptation (CycleGAN and CUT), we also reveal the advantage of diffusion models towards improved segmentation and image translation quality. With a Dice score of 0.78$\pm$0.27, our proposed method outperformed the compared methods, including SynSeg-Net, while providing intuitive uncertainty measures to further facilitate quality control of the automatic segmentation outcomes. The implementation of our proposed method is available at: https://github.com/HealthX-Lab/DiffusionSynCTSeg.

Figures (3)

• Figure 1: Overview of the proposed uncertainty-aware brain ventricle segmentation technique, based on Unpaired Neural Schrödinger Bridge (UNSB) and joint image translation and segmentation training. Note that in the test stage, only the segmentation module is used to obtain segmentation in real CT scans.
• Figure 2: An overview of two different training approaches for ventricle segmentation. Left: Two-stage approach with image translation and segmentation modules trained separately. Right: End-to-end approach with a unified loss function for both tasks.
• Figure 3: Qualitative demonstration of ventricle segmentation and uncertainty measures for two subjects with two different slices using different end-to-end methods.