Multicenter PET image harmonization using generative adversarial networks

Abstract

Abstract Purpose To improve reproducibility and predictive performance of PET radiomic features in multicentric studies by cycle-consistent generative adversarial network (GAN) harmonization approaches. Methods GAN-harmonization was developed to harmonize whole-body PET scans to perform image style and texture translation between different centers and scanners. GAN-harmonization was evaluated by application to two retrospectively collected open datasets and different tasks. First, GAN-harmonization was performed on a dual-center lung cancer cohort (127 female, 138 male) where the reproducibility of radiomic features in healthy liver tissue was evaluated. Second, GAN-harmonization was applied to a head and neck cancer cohort (43 female, 154 male) acquired from three centers. Here, the clinical impact of GAN-harmonization was analyzed by predicting the development of distant metastases using a logistic regression model incorporating first-order statistics and texture features from baseline 18 F-FDG PET before and after harmonization. Results Image quality remained high (structural similarity: left kidney $$\ge$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>≥</mml:mo> </mml:math> 0.800, right kidney $$\ge$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>≥</mml:mo> </mml:math> 0.806, liver $$\ge$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>≥</mml:mo> </mml:math> 0.780, lung $$\ge$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>≥</mml:mo> </mml:math> 0.838, spleen $$\ge$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>≥</mml:mo> </mml:math> 0.793, whole-body $$\ge$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>≥</mml:mo> </mml:math> 0.832) after image harmonization across all utilized datasets. Using GAN-harmonization, inter-site reproducibility of radiomic features in healthy liver tissue increased at least by $$\ge$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>≥</mml:mo> </mml:math> 5 ± 14% (first-order), $$\ge$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>≥</mml:mo> </mml:math> 16 ± 7% (GLCM), $$\ge$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>≥</mml:mo> </mml:math> 19 ± 5% (GLRLM), $$\ge$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>≥</mml:mo> </mml:math> 16 ± 8% (GLSZM), $$\ge$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>≥</mml:mo> </mml:math> 17 ± 6% (GLDM), and $$\ge$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>≥</mml:mo> </mml:math> 23 ± 14% (NGTDM). In the head and neck cancer cohort, the outcome prediction improved from AUC 0.68 (95% CI 0.66–0.71) to AUC 0.73 (0.71–0.75) by application of GAN-harmonization. Conclusions GANs are capable of performing image harmonization and increase reproducibility and predictive performance of radiomic features derived from different centers and scanners.

Topics & Concepts