Impact of harmonization and oversampling methods on radiomics analysis of multi-center imbalanced datasets: application to PET-based prediction of lung cancer subtypes

Abstract Background Medical imaging data frequently encounter image-generation heterogeneity and class imbalance properties, challenging strong generalized predictive performances with data-driven machine-learning methods. The purpose of this study was to investigate the impact of harmonization and...

Full description

Saved in:
Bibliographic Details
Main Authors: Dongyang Du, Isaac Shiri, Fereshteh Yousefirizi, Mohammad R. Salmanpour, Jieqin Lv, Huiqin Wu, Wentao Zhu, Habib Zaidi, Lijun Lu, Arman Rahmim
Format: Article
Language:English
Published: SpringerOpen 2025-04-01
Series:EJNMMI Physics
Subjects:
Online Access:https://doi.org/10.1186/s40658-025-00750-7
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849737609476046848
author Dongyang Du
Isaac Shiri
Fereshteh Yousefirizi
Mohammad R. Salmanpour
Jieqin Lv
Huiqin Wu
Wentao Zhu
Habib Zaidi
Lijun Lu
Arman Rahmim
author_facet Dongyang Du
Isaac Shiri
Fereshteh Yousefirizi
Mohammad R. Salmanpour
Jieqin Lv
Huiqin Wu
Wentao Zhu
Habib Zaidi
Lijun Lu
Arman Rahmim
author_sort Dongyang Du
collection DOAJ
description Abstract Background Medical imaging data frequently encounter image-generation heterogeneity and class imbalance properties, challenging strong generalized predictive performances with data-driven machine-learning methods. The purpose of this study was to investigate the impact of harmonization and oversampling methods on multi-center imbalanced datasets, with specific application to PET-based radiomics modeling for histologic subtype prediction in non-small cell lung cancer (NSCLC). Methods The retrospective study included 245 patients with adenocarcinoma (ADC) and 78 patients with squamous cell carcinoma (SCC) from 4 centers. Utilizing 1502 radiomics features per patient, we trained, validated, and tested 4 machine-learning classifiers, to investigate the effect of no harmonization (NoH) or 4 feature harmonization methods, paired with no oversampling (NoO) or 5 oversampling methods on subtype prediction. Model performance was evaluated using the average area under the ROC curve (AUROC) and G-mean via 5 times 5-fold cross-validations. Statistical comparisons of the combined models against baseline (NoH + NoO) were performed for each fold of cross-validation using the DeLong test. Results The number of cross-combinations with both AUROC and G-mean outperforming baseline in validation and testing was 15, 4, 2, and 7 (out of 29) for random forest (RF), linear discriminant analysis (LDA), logistic regression (LR), and support vector machine (SVM), respectively. ComBat harmonization combined with oversampling (SMOTE) via RF yielded better performance than baseline (AUROC and G-mean of validation: 0.725 vs. 0.608 and 0.625 vs. 0.398; testing: 0.637 vs. 0.567 and 0.506 vs. 0.287), though statistical significances were not observed. Conclusions Applying harmonization and oversampling methods in multi-center imbalanced datasets can improve NSCLC-subtype prediction, but the effect varies widely across classifiers. We have created open-source comparisons of harmonization and oversampling on different classifiers for comprehensive evaluations in different studies.
format Article
id doaj-art-ed62b6da2fa14ea6b14289164fa5bbfd
institution DOAJ
issn 2197-7364
language English
publishDate 2025-04-01
publisher SpringerOpen
record_format Article
series EJNMMI Physics
spelling doaj-art-ed62b6da2fa14ea6b14289164fa5bbfd2025-08-20T03:06:52ZengSpringerOpenEJNMMI Physics2197-73642025-04-0112111610.1186/s40658-025-00750-7Impact of harmonization and oversampling methods on radiomics analysis of multi-center imbalanced datasets: application to PET-based prediction of lung cancer subtypesDongyang Du0Isaac Shiri1Fereshteh Yousefirizi2Mohammad R. Salmanpour3Jieqin Lv4Huiqin Wu5Wentao Zhu6Habib Zaidi7Lijun Lu8Arman Rahmim9College of Computer Science, Inner Mongolia UniversityDivision of Nuclear Medicine and Molecular Imaging, Geneva University HospitalDepartment of Integrative Oncology, BC Cancer Research InstituteDepartment of Integrative Oncology, BC Cancer Research InstituteSchool of Biomedical Engineering, Guangdong Provincial Key Laboratory of Medical Image Processing, Southern Medical UniversitySchool of Biomedical Engineering, Guangdong Provincial Key Laboratory of Medical Image Processing, Southern Medical UniversityResearch Center for Healthcare Data ScienceDivision of Nuclear Medicine and Molecular Imaging, Geneva University HospitalSchool of Biomedical Engineering, Guangdong Provincial Key Laboratory of Medical Image Processing, Southern Medical UniversityDepartment of Integrative Oncology, BC Cancer Research InstituteAbstract Background Medical imaging data frequently encounter image-generation heterogeneity and class imbalance properties, challenging strong generalized predictive performances with data-driven machine-learning methods. The purpose of this study was to investigate the impact of harmonization and oversampling methods on multi-center imbalanced datasets, with specific application to PET-based radiomics modeling for histologic subtype prediction in non-small cell lung cancer (NSCLC). Methods The retrospective study included 245 patients with adenocarcinoma (ADC) and 78 patients with squamous cell carcinoma (SCC) from 4 centers. Utilizing 1502 radiomics features per patient, we trained, validated, and tested 4 machine-learning classifiers, to investigate the effect of no harmonization (NoH) or 4 feature harmonization methods, paired with no oversampling (NoO) or 5 oversampling methods on subtype prediction. Model performance was evaluated using the average area under the ROC curve (AUROC) and G-mean via 5 times 5-fold cross-validations. Statistical comparisons of the combined models against baseline (NoH + NoO) were performed for each fold of cross-validation using the DeLong test. Results The number of cross-combinations with both AUROC and G-mean outperforming baseline in validation and testing was 15, 4, 2, and 7 (out of 29) for random forest (RF), linear discriminant analysis (LDA), logistic regression (LR), and support vector machine (SVM), respectively. ComBat harmonization combined with oversampling (SMOTE) via RF yielded better performance than baseline (AUROC and G-mean of validation: 0.725 vs. 0.608 and 0.625 vs. 0.398; testing: 0.637 vs. 0.567 and 0.506 vs. 0.287), though statistical significances were not observed. Conclusions Applying harmonization and oversampling methods in multi-center imbalanced datasets can improve NSCLC-subtype prediction, but the effect varies widely across classifiers. We have created open-source comparisons of harmonization and oversampling on different classifiers for comprehensive evaluations in different studies.https://doi.org/10.1186/s40658-025-00750-7PET radiomicsHarmonizationOversamplingMulti-center imbalanced datasetsNSCLC
spellingShingle Dongyang Du
Isaac Shiri
Fereshteh Yousefirizi
Mohammad R. Salmanpour
Jieqin Lv
Huiqin Wu
Wentao Zhu
Habib Zaidi
Lijun Lu
Arman Rahmim
Impact of harmonization and oversampling methods on radiomics analysis of multi-center imbalanced datasets: application to PET-based prediction of lung cancer subtypes
EJNMMI Physics
PET radiomics
Harmonization
Oversampling
Multi-center imbalanced datasets
NSCLC
title Impact of harmonization and oversampling methods on radiomics analysis of multi-center imbalanced datasets: application to PET-based prediction of lung cancer subtypes
title_full Impact of harmonization and oversampling methods on radiomics analysis of multi-center imbalanced datasets: application to PET-based prediction of lung cancer subtypes
title_fullStr Impact of harmonization and oversampling methods on radiomics analysis of multi-center imbalanced datasets: application to PET-based prediction of lung cancer subtypes
title_full_unstemmed Impact of harmonization and oversampling methods on radiomics analysis of multi-center imbalanced datasets: application to PET-based prediction of lung cancer subtypes
title_short Impact of harmonization and oversampling methods on radiomics analysis of multi-center imbalanced datasets: application to PET-based prediction of lung cancer subtypes
title_sort impact of harmonization and oversampling methods on radiomics analysis of multi center imbalanced datasets application to pet based prediction of lung cancer subtypes
topic PET radiomics
Harmonization
Oversampling
Multi-center imbalanced datasets
NSCLC
url https://doi.org/10.1186/s40658-025-00750-7
work_keys_str_mv AT dongyangdu impactofharmonizationandoversamplingmethodsonradiomicsanalysisofmulticenterimbalanceddatasetsapplicationtopetbasedpredictionoflungcancersubtypes
AT isaacshiri impactofharmonizationandoversamplingmethodsonradiomicsanalysisofmulticenterimbalanceddatasetsapplicationtopetbasedpredictionoflungcancersubtypes
AT fereshtehyousefirizi impactofharmonizationandoversamplingmethodsonradiomicsanalysisofmulticenterimbalanceddatasetsapplicationtopetbasedpredictionoflungcancersubtypes
AT mohammadrsalmanpour impactofharmonizationandoversamplingmethodsonradiomicsanalysisofmulticenterimbalanceddatasetsapplicationtopetbasedpredictionoflungcancersubtypes
AT jieqinlv impactofharmonizationandoversamplingmethodsonradiomicsanalysisofmulticenterimbalanceddatasetsapplicationtopetbasedpredictionoflungcancersubtypes
AT huiqinwu impactofharmonizationandoversamplingmethodsonradiomicsanalysisofmulticenterimbalanceddatasetsapplicationtopetbasedpredictionoflungcancersubtypes
AT wentaozhu impactofharmonizationandoversamplingmethodsonradiomicsanalysisofmulticenterimbalanceddatasetsapplicationtopetbasedpredictionoflungcancersubtypes
AT habibzaidi impactofharmonizationandoversamplingmethodsonradiomicsanalysisofmulticenterimbalanceddatasetsapplicationtopetbasedpredictionoflungcancersubtypes
AT lijunlu impactofharmonizationandoversamplingmethodsonradiomicsanalysisofmulticenterimbalanceddatasetsapplicationtopetbasedpredictionoflungcancersubtypes
AT armanrahmim impactofharmonizationandoversamplingmethodsonradiomicsanalysisofmulticenterimbalanceddatasetsapplicationtopetbasedpredictionoflungcancersubtypes