The impact of MRI image quality on statistical and predictive analysis of voxel based morphology
Poster No:
1683
Submission Type:
Abstract Submission
Authors:
Felix Hoffstaedter1, Nicolás Nieto2, Simon Eickhoff3, Kaustubh Patil4
Institutions:
1Research Centre Jülich, Jülich, Germany, 2Research Center Jülich, Köln, NRW, 3Research Centre Jülich, Jülich, NRW, 4Research Center Jülich, Jülich, NRW
First Author:
Co-Author(s):
Introduction:
The quality of structural MRI data significantly impacts derivative measures of brain morphology. Within scanner motion, as the most common source of artifacts was shown to reduce both gray matter volume as well as cortical thickness estimates (3). Accurate image quality assessment remains critical for both clinical diagnoses and research, with no universally accepted quality threshold available. Visual image inspection still remains standard procedure, especially for borderline-quality scans, automated quality assurance becomes ever more important with increasing sample sizes. Tools like CAT12 and Freesurfer provide quality estimates that align well with expert ratings (1, 2) and commonly images with severe artifacts are excluded from analysis, but the impact of relative image quality on statistical analysis remains elusive. Here, we investigated the effect of MR image quality on both statistical sex/gender differences in voxel-based morphology (VBM) and on the prediction of sex/gender from VBM in a machine learning framework.
Methods:
We used 5 large datasets of healthy adults aged 18–80: SALD (N=494), eNKI (N=818), CamCAN (N=651), 1000Brains (N=1134), and AOMIC-ID1000 (N=922). Data were pre-processed with CAT12.8.1, generating a comparable image quality rating (IQR) and voxel-wise gray matter volumes in MNI space. IQR scores of successful MRI scans range from 1 (excellent) to <4 (poor). In this study, images feature good to satisfactory quality across datasets with mean IQR ranging from 1.90 (high-quality) to 3.09 (low-quality). Within datasets, high- and low-quality subsamples as well as 20 matching random-quality sets were created based on IQR. Since age is the major source of variance in VBM, we stratified for sex/gender balanced subsamples into 10/3 age bins resulting in 200/336 participants for SALD (median IQR 2.32), 280/426 for eNKI (M IQR 2.19) and 280/322 for CamCAN (M IQR 2.29). To minimize subsample overlap, only 10 age bins were created for AOMIC (n=460) with higher median IQR of 1.98 and 1000Brains (n=220) with lower median IQR of 2.47. For univariate statistics, total intracranial volume (TIV) was linearly regressed before applying t-tests for sex/gender differences (Bonferroni corrected, p<.05) on pooled data and the number of significant tests was recorded per quality subsample. For machine learning, logistic regression was used to predict sex/gender after leakage-free confound regression of TIV from the features. One model was trained for each dataset and sampling strategy and an additional model was trained using the pooled data. To address generalizability, a 5 times repeated 5-fold cross-validation approach was used and the Area under the Receiver Operating Characteristic Curve (AUC) was obtained on the test folds (github.com/N-Nieto/QC).
Results:
In univariate analyses, poorer image quality drastically reduced sensitivity to sex/gender effects. For 10/3 age bins, 1.8%/8.9% of t-tests were significant for low- compared to 7.1%/17.3% for high-quality data and 2.6%/13.4% of positive results for intermediate quality (Fig.1). Sex/gender classification accuracy was very similar ~77% across different qualities for pooled data of 10 age bins as with eNKI and CamCAN subsamples of different quality (Fig.2). Both 1000Brains and SALD datasets with lower overall quality, showed an increase in accuracy from 64% to 71% in SALD and 55% to 62% in 1000brains for higher quality, while high quality AOMIC data even showed a slight decrease from 90% to 87%.
·Fig.1: Number of significant features of pooled data after Bonferroni correction.
·Fig.2: AUC for sex/gender classification for subsamples with 10 age bins for low, random and high-quality images and for the pooled data.
Conclusions:
For univariate analyses, we found that poorer image quality consistently resulted in much lower sensitivity for group differences, which suggests that focusing on higher image quality increases the chance of detecting effects in classical group comparisons. Conversely, classification analysis was much less influenced by image quality in the current setup with generally acceptable image quality suggesting that machine learning models are quite robust to variations in image quality.
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Motion Correction and Preprocessing
Univariate Modeling 1
Other Methods
Keywords:
Machine Learning
STRUCTURAL MRI
Univariate
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Gilmore, A. D., Buser, N. J., & Hanson, J. L. (2021). Variations in structural MRI quality significantly impact commonly used measures of brain anatomy. Brain Informatics, 8(1), 7. https://doi.org/10.1186/s40708-021-00128-2
Reuter, M., Tisdall, M. D., Qureshi, A., Buckner, R. L., Van Der Kouwe, A. J. W., & Fischl, B. (2015). Head motion during MRI acquisition reduces gray matter volume and thickness estimates. NeuroImage, 107, 107–115. https://doi.org/10.1016/j.neuroimage.2014.12.006
Rosen, A. F. G., Roalf, D. R., Ruparel, K., Blake, J., Seelaus, K., Villa, L. P., Ciric, R., Cook, P. A., Davatzikos, C., Elliott, M. A., Garcia De La Garza, A., Gennatas, E. D., Quarmley, M., Schmitt, J. E., Shinohara, R. T., Tisdall, M. D., Craddock, R. C., Gur, R. E., Gur, R. C., & Satterthwaite, T. D. (2018). Quantitative assessment of structural image quality. NeuroImage, 169, 407–418. https://doi.org/10.1016/j.neuroimage.2017.12.059