Re-DiffiNet: Modeling discrepancy in tumor segmentation using diffusion models

Tianyi Ren, Abhishek Sharma, Juampablo E Heras Rivera, Lakshmi Harshitha Rebala, Ethan Honey, Agamdeep Chopra, Mehmet Kurt
Proceedings of The 7nd International Conference on Medical Imaging with Deep Learning, PMLR 250:1257-1266, 2024.

Abstract

Identification of tumor margins is essential for surgical decision-making for glioblastoma patients and provides reliable assistance for neurosurgeons. Despite improvements in deep learning architectures for tumor segmentation over the years, creating a fully autonomous system suitable for clinical floors remains a formidable challenge because the model predictions have not yet reached the desired level of accuracy and generalizability for clinical applications. Generative modeling techniques have seen significant improvements in recent times. Specifically, Generative Adversarial Networks (GANs) and Denoising diffusion probabilistic models (DDPMs) have been used to generate higher-quality images with fewer artifacts and finer attributes. In this work, we introduce a framework called Re-Diffinet for modeling the discrepancy between the outputs of a segmentation model like U-Net and the ground truth, using DDPMs. By explicitly modeling the discrepancy, the results show an average improvement of 0.55% in the Dice score and 16.28% in 95% Hausdorff Distance from cross-validation over 5-folds, compared to the state-of-the-art U-Net segmentation model. The code is available:

Cite this Paper

BibTeX
@InProceedings{pmlr-v250-ren24a, title = {Re-DiffiNet: Modeling discrepancy in tumor segmentation using diffusion models}, author = {Ren, Tianyi and Sharma, Abhishek and Rivera, Juampablo E Heras and Rebala, Lakshmi Harshitha and Honey, Ethan and Chopra, Agamdeep and Kurt, Mehmet}, booktitle = {Proceedings of The 7nd International Conference on Medical Imaging with Deep Learning}, pages = {1257--1266}, year = {2024}, editor = {Burgos, Ninon and Petitjean, Caroline and Vakalopoulou, Maria and Christodoulidis, Stergios and Coupe, Pierrick and Delingette, Hervé and Lartizien, Carole and Mateus, Diana}, volume = {250}, series = {Proceedings of Machine Learning Research}, month = {03--05 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v250/main/assets/ren24a/ren24a.pdf}, url = {https://proceedings.mlr.press/v250/ren24a.html}, abstract = {Identification of tumor margins is essential for surgical decision-making for glioblastoma patients and provides reliable assistance for neurosurgeons. Despite improvements in deep learning architectures for tumor segmentation over the years, creating a fully autonomous system suitable for clinical floors remains a formidable challenge because the model predictions have not yet reached the desired level of accuracy and generalizability for clinical applications. Generative modeling techniques have seen significant improvements in recent times. Specifically, Generative Adversarial Networks (GANs) and Denoising diffusion probabilistic models (DDPMs) have been used to generate higher-quality images with fewer artifacts and finer attributes. In this work, we introduce a framework called Re-Diffinet for modeling the discrepancy between the outputs of a segmentation model like U-Net and the ground truth, using DDPMs. By explicitly modeling the discrepancy, the results show an average improvement of 0.55% in the Dice score and 16.28% in 95% Hausdorff Distance from cross-validation over 5-folds, compared to the state-of-the-art U-Net segmentation model. The code is available:} }
Endnote
%0 Conference Paper %T Re-DiffiNet: Modeling discrepancy in tumor segmentation using diffusion models %A Tianyi Ren %A Abhishek Sharma %A Juampablo E Heras Rivera %A Lakshmi Harshitha Rebala %A Ethan Honey %A Agamdeep Chopra %A Mehmet Kurt %B Proceedings of The 7nd International Conference on Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2024 %E Ninon Burgos %E Caroline Petitjean %E Maria Vakalopoulou %E Stergios Christodoulidis %E Pierrick Coupe %E Hervé Delingette %E Carole Lartizien %E Diana Mateus %F pmlr-v250-ren24a %I PMLR %P 1257--1266 %U https://proceedings.mlr.press/v250/ren24a.html %V 250 %X Identification of tumor margins is essential for surgical decision-making for glioblastoma patients and provides reliable assistance for neurosurgeons. Despite improvements in deep learning architectures for tumor segmentation over the years, creating a fully autonomous system suitable for clinical floors remains a formidable challenge because the model predictions have not yet reached the desired level of accuracy and generalizability for clinical applications. Generative modeling techniques have seen significant improvements in recent times. Specifically, Generative Adversarial Networks (GANs) and Denoising diffusion probabilistic models (DDPMs) have been used to generate higher-quality images with fewer artifacts and finer attributes. In this work, we introduce a framework called Re-Diffinet for modeling the discrepancy between the outputs of a segmentation model like U-Net and the ground truth, using DDPMs. By explicitly modeling the discrepancy, the results show an average improvement of 0.55% in the Dice score and 16.28% in 95% Hausdorff Distance from cross-validation over 5-folds, compared to the state-of-the-art U-Net segmentation model. The code is available:
APA
Ren, T., Sharma, A., Rivera, J.E.H., Rebala, L.H., Honey, E., Chopra, A. & Kurt, M.. (2024). Re-DiffiNet: Modeling discrepancy in tumor segmentation using diffusion models. Proceedings of The 7nd International Conference on Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 250:1257-1266 Available from https://proceedings.mlr.press/v250/ren24a.html.

Related Material