A novel approach to harmonize MRI-derived morphological features between different MRI scanners
Poster No:
2249
Submission Type:
Abstract Submission
Authors:
Yoonho Hwang1, Min-Hee Lee2, Seung Ku Lee1, Soriul Soriul Kim1, Ali Siddiquee1, Hyeon Jin Kim3, Peter Hadar4, M. Brandon Westover5, Robert Thomas5, Chol Shin1
Institutions:
1Korea University, Ansan, South Korea, 2Wayne State University, Detroit, MI, 3Korea University Ansan Hospital, Ansan, South Korea, 4Massachusetts General Hospital, Boston, MA, 5Beth Israel Deaconess Medical Center, Boston, MA
First Author:
Co-Author(s):
Introduction:
There are increasing efforts to combine and share a huge neuroimaging dataset, especially across large-scale cohorts from multicenter [1,2]. Constructing the dataset is important because it allows the creation of analyzable datasets for more targeted study designs and enables multicenter collaborations [3]. However, when combining the dataset under various environment, we are inevitably faced with the challenge of dealing with various conditions, such as scanner, site, and acquisition parameters that can impact downstream analysis (i.e., extract cortical thickness, etc.) [4]. This effect, called the scanner (or site) effect in neuroimaging, must be removed for consistency and reproducibility of the analysis [5]. This study aims to test the comparability and calibrate the structural MRI-derived morphological features obtained from different MRI scanners.
Methods:
T1-weigted (T1w) images were obtained from 48 participants (52.16±11.3 years old; age range: 31-75 years old; 24 men) who underwent 1.5T and 3T MRI scans within 2 weeks. The 1.5T T1w images were acquired on a 1.5T GE SIGNA scanner at Central Hospital, and the 3T T1w images were acquired on a 3T Siemens Skyra scanner at Korea University Ansan Hospital. The acquisition parameters of the 3D T1w images were as follows: 1) GE 1.5T MRI scanner with a fast spoiled gradient echo sequence (TR = 10.9ms, TE = 5.01ms, TI = 450ms, flip angle = 13◦, voxel size = 0.47 x 0.47 x 1.2 mm3) and Siemens 3T one with a magnetization-prepared rapid gradient-echo sequence (TR = 1980ms, TE = 2.55ms, TI = 998ms, flip angle = 9◦, voxel size = 0.54 x 0.54 x 0.9 mm3). Before the surface-based analysis using FreeSurfer 7.3.2 [6,7], an automated segmentation software, all T1w images were processed using SynthSR that is one of the artificial intelligence techniques to turn heterogeneous MRI scans into high-resolution T1w images and convert to 1mm MPRAGE images [8]. The SynthSR-processed images were analyzed using the 'recon-all' pipeline of FreeSurfer. A visual inspection was performed through Freeview to ensure that the skull was properly removed with the cerebellum, that the pial surface and white matter surface were well defined, and that the brain regions were well segmented. Each cortical thickness (CTh) was extracted from 68 regions based on the Desikan-Killiany atlas [9] using aparcstats2table function. ComBat [4,5], which is an extended linear model for adjusting residuals based on the empirical Bayes to remove additive and multiplicative effects by scanners (or sites), was applied to harmonize the extracted CThs of 1.5T and 3T preserving the interested covariates (i.e., age and gender). Paired t-test was used for comparison of each brain regions between 1.5T and 3T with Bonferroni correction and Bland–Altman plot was used in analyzing the agreement between two different data.
Results:
Figure 1 and 2 show the results of paired t-test and Bland–Altman plot for areas that can be greatly affected in extracting morphological features among 68 regions in the brain before/after using ComBat. As shown in Figure 1, before using ComBat, there was a statistically significant difference in CTh between 1.5T and 3T in the 8 regions related to OSA, whereas after using ComBat, there was no statistically significant difference in the same regions. Figure 2 represents graphically little difference in average CTh after using ComBat compared to before using ComBat, demonstrating the agreement of CTh between 1.5T and 3T. Additionally, the remaining 60 areas denoted the same results as above after using ComBat.
Conclusions:
Our findings suggest that the proposed approach could minimize scanner-specific difference in the CTh of 1.5T and 3T in all segmented areas. It can be applied to horizontal analyses of big data obtained from multicenter studies.
Modeling and Analysis Methods:
Bayesian Modeling
Multivariate Approaches 2
Neuroinformatics and Data Sharing:
Informatics Other 1
Keywords:
Cortex
Data analysis
Open-Source Software
STRUCTURAL MRI
Other - Harmonization, ComBat, Cortical thickness, FreeSurfer, SynthSR
1|2Indicates the priority used for review
Provide references using author date format
2.Petersen,R.C., Aisen,P.S., Beckett,L.A., Donohue,M.C., Gamst,A.C., Harvey,D.J., Jack,J.C.R., Jagust,W.J., Shaw,L.M., Toga,A.W., Trojanowski,J.Q., and Weiner,M.W. (2010), 'AlzheimerDisease Neuroimaging Initiative (ADNI)', Neurology, vol. 74, no. 3, pp. 201–209.
3.Torbati,M.E., Minhas,D.S., Ahmad,G., O’Connor,E.E., Muschelli,J., Laymon,C.M., Yang,Z., Cohen,A.D., Aizenstein,H.J., Klunk,W.E., Christian,B.T., Hwang,S.J., Crainiceanu,C.M., and Tudorascu,D.L. (2021), 'A multi-scanner neuroimaging data harmonization using RAVEL and ComBat', NeuroImage, vol. 245, pp. 118703.
4.Fortin,J.P., Cullen,N., Sheline,Y.I., Taylor,W.D., Aselcioglu,I., Cook,P.A., Adams,P., Cooper,C., Fava,M., McGrath,P.J. and McInnis,M. (2018), 'Harmonization of cortical thickness measurements across scanners and sites', Neuroimage, vol. 167, pp. 104-120.
5.Johnson,W.E., Li,C., and Rabinovic,A. (2007), 'Adjusting batch effects in microarray expression data using empirical Bayes methods', Biostatistics, vol. 8, no. 1, pp. 118–127.
6.Fischl,B., Salat,D.H., Busa,E., Albert,M., Dieterich,M., Haselgrove,C., van der Kouwe,A., Killiany,R., Kennedy,D., Klaveness,S., Montillo,A., Makris,N., Rosen,B., and Dale,A.M. (2002), 'Whole Brain Segmentation: Automated Labeling of Neuroanatomical Structures in the Human Brain', Neuron, vol. 33, no. 3, pp. 341–355.
7.Fischl,B., Salat,D.H., van der Kouwe,A.J.W., Makris,N., Ségonne,F., Quinn,B.T., and Dale,A.M. (2004), 'Sequence-independent segmentation of magnetic resonance images', NeuroImage, vol. 23, pp. 69–84.
8.Iglesias,J.E., Billot,B., Balbastre,Y., Magdamo,C., Arnold,S.E., Das,S., Edlow,B.L., Alexander,D.C., Golland,P. and Fischl,B. (2023), 'SynthSR: A public AI tool to turn heterogeneous clinical brain scans into high-resolution T1-weighted images for 3D morphometry', Science advances, vol. 9, no. 5, pp. eadd3607.
9.Desikan,R.S., Ségonne,F., Fischl,B., Quinn,B.T., Dickerson,B.C., Blacker,D., Buckner,R.L., Dale,A.M., Maguire,R.P., Hyman,B.T., Albert,M.S., and Killiany,R.J. (2006), 'An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest', NeuroImage, vol. 31, no. 3, pp. 968–980.