Efficacy of MRI data harmonization in the age of machine learning: a multicenter study across 36 datasets
Abstract Pooling publicly-available MRI data from multiple sites allows to assemble extensive groups of subjects, increase statistical power, and promote data reuse with machine learning techniques. The harmonization of multicenter data is necessary to reduce the confounding effect associated with n...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2024-01-01
|
| Series: | Scientific Data |
| Online Access: | https://doi.org/10.1038/s41597-023-02421-7 |
| _version_ | 1797340349285269504 |
|---|---|
| author | Chiara Marzi Marco Giannelli Andrea Barucci Carlo Tessa Mario Mascalchi Stefano Diciotti |
| author_facet | Chiara Marzi Marco Giannelli Andrea Barucci Carlo Tessa Mario Mascalchi Stefano Diciotti |
| author_sort | Chiara Marzi |
| collection | DOAJ |
| description | Abstract Pooling publicly-available MRI data from multiple sites allows to assemble extensive groups of subjects, increase statistical power, and promote data reuse with machine learning techniques. The harmonization of multicenter data is necessary to reduce the confounding effect associated with non-biological sources of variability in the data. However, when applied to the entire dataset before machine learning, the harmonization leads to data leakage, because information outside the training set may affect model building, and potentially falsely overestimate performance. We propose a 1) measurement of the efficacy of data harmonization; 2) harmonizer transformer, i.e., an implementation of the ComBat harmonization allowing its encapsulation among the preprocessing steps of a machine learning pipeline, avoiding data leakage by design. We tested these tools using brain T1-weighted MRI data from 1740 healthy subjects acquired at 36 sites. After harmonization, the site effect was removed or reduced, and we showed the data leakage effect in predicting individual age from MRI data, highlighting that introducing the harmonizer transformer into a machine learning pipeline allows for avoiding data leakage by design. |
| first_indexed | 2024-03-08T10:00:49Z |
| format | Article |
| id | doaj.art-64e6ed0b46e1461ea042fc07c77933f5 |
| institution | Directory Open Access Journal |
| issn | 2052-4463 |
| language | English |
| last_indexed | 2024-03-08T10:00:49Z |
| publishDate | 2024-01-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | Scientific Data |
| spelling | doaj.art-64e6ed0b46e1461ea042fc07c77933f52024-01-29T10:56:48ZengNature PortfolioScientific Data2052-44632024-01-0111112710.1038/s41597-023-02421-7Efficacy of MRI data harmonization in the age of machine learning: a multicenter study across 36 datasetsChiara Marzi0Marco Giannelli1Andrea Barucci2Carlo Tessa3Mario Mascalchi4Stefano Diciotti5Department of Statistics, Computer Science and Applications “Giuseppe Parenti”, University of FlorenceUnit of Medical Physics, Pisa University Hospital “Azienda Ospedaliero-Universitaria Pisana”“Nello Carrara” Institute of Applied Physics (IFAC), National Research Council (CNR)Radiology Unit Apuane e Lunigiana, Azienda USL Toscana Nord OvestDepartment of Experimental and Clinical Biomedical Sciences “Mario Serio”, University of FlorenceDepartment of Electrical, Electronic, and Information Engineering “Guglielmo Marconi” - DEI, University of BolognaAbstract Pooling publicly-available MRI data from multiple sites allows to assemble extensive groups of subjects, increase statistical power, and promote data reuse with machine learning techniques. The harmonization of multicenter data is necessary to reduce the confounding effect associated with non-biological sources of variability in the data. However, when applied to the entire dataset before machine learning, the harmonization leads to data leakage, because information outside the training set may affect model building, and potentially falsely overestimate performance. We propose a 1) measurement of the efficacy of data harmonization; 2) harmonizer transformer, i.e., an implementation of the ComBat harmonization allowing its encapsulation among the preprocessing steps of a machine learning pipeline, avoiding data leakage by design. We tested these tools using brain T1-weighted MRI data from 1740 healthy subjects acquired at 36 sites. After harmonization, the site effect was removed or reduced, and we showed the data leakage effect in predicting individual age from MRI data, highlighting that introducing the harmonizer transformer into a machine learning pipeline allows for avoiding data leakage by design.https://doi.org/10.1038/s41597-023-02421-7 |
| spellingShingle | Chiara Marzi Marco Giannelli Andrea Barucci Carlo Tessa Mario Mascalchi Stefano Diciotti Efficacy of MRI data harmonization in the age of machine learning: a multicenter study across 36 datasets Scientific Data |
| title | Efficacy of MRI data harmonization in the age of machine learning: a multicenter study across 36 datasets |
| title_full | Efficacy of MRI data harmonization in the age of machine learning: a multicenter study across 36 datasets |
| title_fullStr | Efficacy of MRI data harmonization in the age of machine learning: a multicenter study across 36 datasets |
| title_full_unstemmed | Efficacy of MRI data harmonization in the age of machine learning: a multicenter study across 36 datasets |
| title_short | Efficacy of MRI data harmonization in the age of machine learning: a multicenter study across 36 datasets |
| title_sort | efficacy of mri data harmonization in the age of machine learning a multicenter study across 36 datasets |
| url | https://doi.org/10.1038/s41597-023-02421-7 |
| work_keys_str_mv | AT chiaramarzi efficacyofmridataharmonizationintheageofmachinelearningamulticenterstudyacross36datasets AT marcogiannelli efficacyofmridataharmonizationintheageofmachinelearningamulticenterstudyacross36datasets AT andreabarucci efficacyofmridataharmonizationintheageofmachinelearningamulticenterstudyacross36datasets AT carlotessa efficacyofmridataharmonizationintheageofmachinelearningamulticenterstudyacross36datasets AT mariomascalchi efficacyofmridataharmonizationintheageofmachinelearningamulticenterstudyacross36datasets AT stefanodiciotti efficacyofmridataharmonizationintheageofmachinelearningamulticenterstudyacross36datasets |