Parameters from site classification to harmonize MRI clinical studies: Application to a multi‐site Parkinson's disease dataset

Abstract

Multi-site MRI datasets are crucial for big data research. However, neuroimaging studies must face the batch effect. Here, we propose an approach that uses the predictive probabilities provided by Gaussian processes (GPs) to harmonize clinicalbased studies. A multi-site dataset of 216 Parkinson's disease (PD) patients and 87 healthy subjects (HS) was used. We performed a site GP classification using MRI data. The outcomes estimated from this classification, redefined like Weighted HAR-Monization PArameters (WHARMPA), were used as regressors in two different clinical studies: A PD versus HS machine learning classification using GP, and a VBM comparison (FWE-p < .05, k = 100). Same studies were also conducted using conventional Boolean site covariates, and without information about site belonging. The results from site GP classification provided high scores, balanced accuracy (BAC) was 98.39% for grey matter images. PD versus HS classification performed better when the WHARMPA were used to harmonize (BAC = 78.60%; AUC = 0.90) than when using the Boolean site information (BAC = 56.31%; AUC = 0.71) and without it (BAC = 57.22%; AUC = 0.73). The VBM analysis harmonized using WHARMPA provided larger and more statistically robust clusters in regions previously reported in PD than when the Boolean site covariates or no corrections were added to the model. In conclusion, WHARMPA might encode global site-effects quantitatively and allow the harmonization of data. This method is user-friendly and provides a powerful solution, without complex implementations, to clean the analyses by removing variability associated with the differences between sites.

Multi-site harmonization of MRI data uncovers machine-learning discrimination capability in barely separable populations: An example from the ABIDE dataset

“…Wang et al ( Wang et al, 2022 ) developed a novel deep-learning domain adaptation framework to tackle the confounding effects for both Alzheimer’s disease and Schizophrenia classification tasks by using the whole minimally preprocessed 3D T1-weighted brain MRI scans of the subjects. Monte-Rubio et al (C. Monte-Rubio et al, 2022 ) proposed an approach using the predictive probabilities pro- vided by Gaussian processes to harmonize multi-site T1-weighted MRI data for Parkinson’s disease classification. Although the latter two methodologies cannot be applied to data extracted from preprocessed images such as cortical and subcortical features, the authors highlighted that harmo- nization is a crucial preprocessing step to be performed before any clinical classification task.…”

Machine Learning for Brain MRI Data Harmonisation: A Systematic Review

“…For implicit approaches, studies have consistently demonstrated that incorporating harmonisation leads to better performance compared to baseline experiments which do not apply harmonisation. Transfer learning approaches tend to outperform statistical covariate techniques [ 50 , 52 , 55 , 56 , 59 , 60 , 61 , 63 , 64 , 65 , 69 , 71 , 74 ], and some studies suggest that adversarial transfer learning methods outperform non-adversarial methods [ 51 , 53 , 55 ]. Multi-task learning is superior to single-task learning in classification tasks [ 75 ].…”

A multi-site study on sex differences in cortical thickness in non-demented Parkinson’s disease

“…We used multi-site MRI data from four research centers: the University of Deusto (Bilbao, Spain; Site 1), the University of Barcelona (Barcelona, Spain; Site 2), the Center of Addiction and Mental Health (CAMH; Toronto, Canada; Site 3), and the University of Cologne (Cologne, Germany; Site 4). The initial sample comprised 216 PD and 87 HC individuals, previously described in Monté-Rubio et al 31 . The PD patients included in this sample fulfilled the UK PD Society Brain Bank diagnostic criteria for PD and were classified as non-demented according to the Level I for PD dementia diagnosis from the Movement Disorder Society Task Force on Dementia in Parkinson’s Disease 32 , 33 .…”