Anatomy-Guided Multi-Path CycleGAN for Lung CT Kernel Harmonization

Aravind Krishnan, Thomas Li, Lucas Walker Remedios, Kaiwen Xu, Lianrui Zuo, Kim L. Sandler, Fabien Maldonado, Bennett Allan Landman
Proceedings of The 8th International Conference on Medical Imaging with Deep Learning, PMLR 301:787-802, 2026.

Abstract

Accurate quantitative measurement in lung computed tomography (CT) imaging often relies on consistent kernel reconstruction across scanners and manufacturers. Harmonization can reduce measurement variability caused by heterogeneous reconstruction kernels; however, harmonization across different manufacturers and scanners remains challenging due to significant differences in reconstruction protocol and positional alignment of subjects, often resulting in anatomical hallucinations. To address this, we propose a multi-path cycleGAN framework that incorporates multi-region anatomical labels and a tissue statistic loss as anatomical regularization to preserve structural integrity during harmonization. We trained our model on 100 scans each of four representative reconstruction kernels from the National Lung Screening Trial (NLST) dataset and evaluated it on 240 withheld scans. Experimental results demonstrate superior performance of our method in both within manufacturer harmonization and cross-manufacture harmonization: Harmonizing hard-to-soft kernel images within a single manufacturer significantly reduces emphysema measurement discrepancies (p < 0.05). Across manufacturers, harmonizing all kernels to a reference soft kernel yields consistent emphysema quantification (p > 0.05) and preserves anatomical structures, as demonstrated by improved Dice similarity coefficient in skeletal muscle and subcutaneous adipose tissue between harmonized and unharmonized images. These findings demonstrate that segmentation-driven anatomical regularization effectively addresses cross-manufacturer discrepancies, ensuring robust quantitative imaging. We release ourcode and model at https://github.com/MASILab/AnatomyconstrainedMultipathGAN.

Cite this Paper

BibTeX
@InProceedings{pmlr-v301-krishnan26a, title = {Anatomy-Guided Multi-Path CycleGAN for Lung CT Kernel Harmonization}, author = {Krishnan, Aravind and Li, Thomas and Remedios, Lucas Walker and Xu, Kaiwen and Zuo, Lianrui and Sandler, Kim L. and Maldonado, Fabien and Landman, Bennett Allan}, booktitle = {Proceedings of The 8th International Conference on Medical Imaging with Deep Learning}, pages = {787--802}, year = {2026}, editor = {Tasdizen, Tolga and Elhabian, Shireen and Summers, Ronald and Chen, Chen and Koch, Lisa and Zhuang, Yan}, volume = {301}, series = {Proceedings of Machine Learning Research}, month = {09--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v301/main/assets/krishnan26a/krishnan26a.pdf}, url = {https://proceedings.mlr.press/v301/krishnan26a.html}, abstract = {Accurate quantitative measurement in lung computed tomography (CT) imaging often relies on consistent kernel reconstruction across scanners and manufacturers. Harmonization can reduce measurement variability caused by heterogeneous reconstruction kernels; however, harmonization across different manufacturers and scanners remains challenging due to significant differences in reconstruction protocol and positional alignment of subjects, often resulting in anatomical hallucinations. To address this, we propose a multi-path cycleGAN framework that incorporates multi-region anatomical labels and a tissue statistic loss as anatomical regularization to preserve structural integrity during harmonization. We trained our model on 100 scans each of four representative reconstruction kernels from the National Lung Screening Trial (NLST) dataset and evaluated it on 240 withheld scans. Experimental results demonstrate superior performance of our method in both within manufacturer harmonization and cross-manufacture harmonization: Harmonizing hard-to-soft kernel images within a single manufacturer significantly reduces emphysema measurement discrepancies (p < 0.05). Across manufacturers, harmonizing all kernels to a reference soft kernel yields consistent emphysema quantification (p > 0.05) and preserves anatomical structures, as demonstrated by improved Dice similarity coefficient in skeletal muscle and subcutaneous adipose tissue between harmonized and unharmonized images. These findings demonstrate that segmentation-driven anatomical regularization effectively addresses cross-manufacturer discrepancies, ensuring robust quantitative imaging. We release ourcode and model at https://github.com/MASILab/AnatomyconstrainedMultipathGAN.} }
Endnote
%0 Conference Paper %T Anatomy-Guided Multi-Path CycleGAN for Lung CT Kernel Harmonization %A Aravind Krishnan %A Thomas Li %A Lucas Walker Remedios %A Kaiwen Xu %A Lianrui Zuo %A Kim L. Sandler %A Fabien Maldonado %A Bennett Allan Landman %B Proceedings of The 8th International Conference on Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2026 %E Tolga Tasdizen %E Shireen Elhabian %E Ronald Summers %E Chen Chen %E Lisa Koch %E Yan Zhuang %F pmlr-v301-krishnan26a %I PMLR %P 787--802 %U https://proceedings.mlr.press/v301/krishnan26a.html %V 301 %X Accurate quantitative measurement in lung computed tomography (CT) imaging often relies on consistent kernel reconstruction across scanners and manufacturers. Harmonization can reduce measurement variability caused by heterogeneous reconstruction kernels; however, harmonization across different manufacturers and scanners remains challenging due to significant differences in reconstruction protocol and positional alignment of subjects, often resulting in anatomical hallucinations. To address this, we propose a multi-path cycleGAN framework that incorporates multi-region anatomical labels and a tissue statistic loss as anatomical regularization to preserve structural integrity during harmonization. We trained our model on 100 scans each of four representative reconstruction kernels from the National Lung Screening Trial (NLST) dataset and evaluated it on 240 withheld scans. Experimental results demonstrate superior performance of our method in both within manufacturer harmonization and cross-manufacture harmonization: Harmonizing hard-to-soft kernel images within a single manufacturer significantly reduces emphysema measurement discrepancies (p < 0.05). Across manufacturers, harmonizing all kernels to a reference soft kernel yields consistent emphysema quantification (p > 0.05) and preserves anatomical structures, as demonstrated by improved Dice similarity coefficient in skeletal muscle and subcutaneous adipose tissue between harmonized and unharmonized images. These findings demonstrate that segmentation-driven anatomical regularization effectively addresses cross-manufacturer discrepancies, ensuring robust quantitative imaging. We release ourcode and model at https://github.com/MASILab/AnatomyconstrainedMultipathGAN.
APA
Krishnan, A., Li, T., Remedios, L.W., Xu, K., Zuo, L., Sandler, K.L., Maldonado, F. & Landman, B.A.. (2026). Anatomy-Guided Multi-Path CycleGAN for Lung CT Kernel Harmonization. Proceedings of The 8th International Conference on Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 301:787-802 Available from https://proceedings.mlr.press/v301/krishnan26a.html.

Related Material