{"id":4228,"date":"2025-12-05T12:08:15","date_gmt":"2025-12-05T17:08:15","guid":{"rendered":"https:\/\/my.vanderbilt.edu\/masi\/?p=4228"},"modified":"2025-12-05T12:20:14","modified_gmt":"2025-12-05T17:20:14","slug":"anatomy-guided-multi-path-cyclegan-for-lung-ct-kernel-harmonization","status":"publish","type":"post","link":"https:\/\/my.vanderbilt.edu\/masi\/2025\/12\/anatomy-guided-multi-path-cyclegan-for-lung-ct-kernel-harmonization\/","title":{"rendered":"Anatomy-Guided Multi-Path CycleGAN for Lung CT Kernel Harmonization"},"content":{"rendered":"

Aravind Krishnan, Thomas Li, Lucas Walker Remedios, Kaiwen Xu, Lianrui Zuo, Kim L Sandler, Fabien Maldonado, Bennett Allan Landman, MIDL 2025, Link to paper: https:\/\/openreview.net\/pdf?id=w3p7GddsQ8<\/a><\/p>\n

 <\/p>\n

Abstract\u00a0<\/strong><\/h2>\n

Accurate quantitative measurement in lung computed tomography (CT) imaging often re-<\/p>\n

lies on consistent kernel reconstruction across scanners and manufacturers. Harmonization<\/p>\n

can reduce measurement variability caused by heterogeneous reconstruction kernels; how-<\/p>\n

ever, harmonization across different manufacturers and scanners remains challenging due<\/p>\n

to significant differences in reconstruction protocol and positional alignment of subjects,<\/p>\n

often resulting in anatomical hallucinations. To address this, we propose a multi-path cy-<\/p>\n

cleGAN framework that incorporates multi-region anatomical labels and a tissue statistic<\/p>\n

loss as anatomical regularization to preserve structural integrity during harmonization. We<\/p>\n

trained our model on 100 scans each of four representative reconstruction kernels from the<\/p>\n

National Lung Screening Trial (NLST) dataset and evaluated it on 240 withheld scans.<\/p>\n

Experimental results demonstrate superior performance of our method in both within-<\/p>\n

manufacturer harmonization and cross-manufacture harmonization: Harmonizing hard-to<\/p>\n

soft-kernel images within a single manufacturer significantly reduces emphysema measure-<\/p>\n

ment discrepancies (p < 0.05). Across manufacturers, harmonizing all kernels to a reference<\/p>\n

soft kernel yields consistent emphysema quantification (p > 0.05) and preserves anatomical<\/p>\n

structures, as demonstrated by improved Dice similarity coefficient in skeletal muscle and<\/p>\n

subcutaneous adipose tissue between harmonized and unharmonized images. These find-<\/p>\n

ings demonstrate that segmentation-driven anatomical regularization effectively addresses<\/p>\n

cross-manufacturer discrepancies, ensuring robust quantitative imaging. We release our<\/p>\n

code and model at <\/span>https:\/\/github.com\/MASILab\/AnatomyconstrainedMultipathGAN.<\/span><\/p>\n

Keywords:<\/strong> cycleGAN, harmonization, CT, emphysema, synthesis<\/p>\n

 <\/p>\n

\"\"<\/p>\n

Figure:<\/strong> For any given pair of reconstruction kernels, the generator is a ResNet, formed<\/p>\n

by a source encoder and target decoder in the forward path and a target encoder<\/p>\n

and source decoder in the backward path. Each generator produces a synthetic<\/p>\n

image with the style of either the source or target kernel. A PatchGAN is used<\/p>\n

as a discriminator for the corresponding domain to distinguish between real and<\/p>\n

synthetic images. The mean of all unique labels are computed using the multilabel<\/p>\n

masks for the real and synthetic image and are penalized such that the anatomy<\/p>\n

remains preserved in the harmonized image.<\/p>\n","protected":false},"excerpt":{"rendered":"

Aravind Krishnan, Thomas Li, Lucas Walker Remedios, Kaiwen Xu, Lianrui Zuo, Kim L Sandler, Fabien Maldonado, Bennett Allan Landman, MIDL 2025, Link to paper: https:\/\/openreview.net\/pdf?id=w3p7GddsQ8   Abstract\u00a0 Accurate quantitative measurement in lung computed tomography (CT) imaging often re- lies on consistent kernel reconstruction across scanners and manufacturers. Harmonization can reduce measurement variability caused by heterogeneous…<\/p>\n","protected":false},"author":9571,"featured_media":4217,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[60,130,131,132,116,209,147],"tags":[],"class_list":["post-4228","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-computed-tomography","category-deep-learning","category-generative-adversarial-networks","category-harmonization","category-image-processing","category-lung-cancer","category-lung-screening-ct"],"_links":{"self":[{"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/posts\/4228","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/users\/9571"}],"replies":[{"embeddable":true,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/comments?post=4228"}],"version-history":[{"count":2,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/posts\/4228\/revisions"}],"predecessor-version":[{"id":4230,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/posts\/4228\/revisions\/4230"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/media\/4217"}],"wp:attachment":[{"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/media?parent=4228"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/categories?post=4228"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/tags?post=4228"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}