Abstract
We propose a complementary feature integration framework for unsupervised 3D medical image segmentation that repurposes a pretrained 3D latent diffusion model as a feature extractor. The framework integrates raw image intensity with multiple representations extracted from different components of the diffusion model, including VAE latent representations, diffusion U-Net features, and self-attention features. These representations are integrated into a unified feature space together with spatial coordinates and subsequently segmented using recursive 3D normalized cuts. We evaluate the proposed framework on BraTS2020 brain tumour MRI and FLARE22 abdominal CT datasets using identical hyperparameters without dataset-specific tuning. Experimental results show that integrating multiple representations improves segmentation performance across both datasets compared with individual representations and baseline methods. The proposed framework achieves the best overall average performance across the evaluated datasets while demonstrating the ability to generalise across MRI and CT modalities. These findings suggest that representations derived from different components of the diffusion framework capture complementary anatomical information that benefits unsupervised 3D medical image segmentation. Code is publicly available at https://doi.org/10.5281/zenodo.21375116.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/DGM4MICCAI_020.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/DGM4MICCAI_020_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/profile?id=~Tanatta_Chaichakan1
BibTex
@InProceedings{ChaTan_Leveraging_MICCAISAT2026,
author = { Chaichakan, Tanatta AND Deligianni, Fani AND Henderson, Paul},
title = { { Leveraging Multi-Representation Features from Diffusion Models for Unsupervised 3D Medical Image Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17258},
month = {pending},
page = {pending}
}