Age dependent volume estimation biases in recon-all clinical
Poster No:
1509
Submission Type:
Abstract Submission
Authors:
Dylan Nielson1, John Lee2, Eric Earl3, Dustin Moraczewski3, Francisco Pereira3
Institutions:
1National Institute of Mental Health, Washington, DC, 2Python AI Solutions, Dublin, Dublin, 3National Institute of Mental Health, Bethesda, MD
First Author:
Co-Author(s):
Introduction:
Cortical morphometry is an essential part of understanding the human brain and, in particular, neurodevelopment, and neurodevelopmental disorders. Estimating cortical morphometry typically requires at least a 1 mm isotropic T1w image and a lengthy computation. However, recently developed deep learning methods may be able to quickly and reliably estimate morphometrics from scans of any resolution or contrast. Freesurfer 7.4.0 made these tools broadly accessible with the recon-all clinical pipeline, trained on 1,000 adults from HCP and ADNI (Gopinath et al, 2024), and the Freesurfer 8.0.0 beta makes them the default approach for estimating cortical morphometry.
This capability is potentially groundbreaking for the study of neurodevelopmental disorders. Patients with these disorders are frequently assessed with MRI during the diagnostic process. If cortical morphometry can be estimated from these lower resolution clinical scans, then much more data becomes available for understanding how these disorders alter neurodevelopmental trajectories. We sought to evaluate the suitability of recon-all clinical for use in children and adolescents by comparing morphometric estimates from recon-all clinical (RAC) to recon-all (RA) estimates in scans from the ABCD study.
This capability is potentially groundbreaking for the study of neurodevelopmental disorders. Patients with these disorders are frequently assessed with MRI during the diagnostic process. If cortical morphometry can be estimated from these lower resolution clinical scans, then much more data becomes available for understanding how these disorders alter neurodevelopmental trajectories. We sought to evaluate the suitability of recon-all clinical for use in children and adolescents by comparing morphometric estimates from recon-all clinical (RAC) to recon-all (RA) estimates in scans from the ABCD study.
Methods:
We randomly selected 1000 sessions from the Adolescent Brain and Cognitive Development (ABCD) sessions in which one T1w and one T2w image passed ABCD's quality control checks. We selected one session per participant per family balanced across the 1st to 4th study waves. We ran RA from Freesurfer 7.4.1 with both the T1w and T2w images as input. We then ran RAC with three different inputs:
- T1w images in their original 1 mm isotropic resolution (RACT1)
- T1w images downsampled to 1 mm x 1 mm x 5 mm (RACT1-R5), to simulate typically thicker clinical slices
- T2w images in their original 1 mm isotropic resolution (RACT2)
All 4 pipelines ran successfully on 993 participants, aged 9 to 18 (median age = 13.08, IQR = 4.16, 463 female, 5 other or unknown).
- T1w images in their original 1 mm isotropic resolution (RACT1)
- T1w images downsampled to 1 mm x 1 mm x 5 mm (RACT1-R5), to simulate typically thicker clinical slices
- T2w images in their original 1 mm isotropic resolution (RACT2)
All 4 pipelines ran successfully on 993 participants, aged 9 to 18 (median age = 13.08, IQR = 4.16, 463 female, 5 other or unknown).
Results:
Correlations between RA and RAC estimates for cortical and subcortical gray matter volumes (GMV) and white matter volume (WMV) are high for all RAC inputs (min r2=0.9). However, RAC estimates for cortical GMV and cerebral WMV show an age dependent bias, with GMV underestimated and WMV overestimated for younger participants, in both RACT1 and RACT1R5 (plotted in Fig. 1A, model coefficients in Fig. 1B). The relationship between WMV error and age is flipped with RACT2. We examined the regional pattern of the age dependence of the error in RACT1 GMV estimates (Fig. 1C) and found that they are inversely correlated with the age at which GMV peaks in each region (left hemisphere: r=-0.59, p=0.00028, right hemisphere: r=-0.43, p=0.012).
Despite this bias, RAC estimates show similar developmental trajectories within this age range for overall tissue volumes (Fig. 2A) as well as for regional estimates of gray matter volume (Fig. 2B; min r2=0.78). We fit a robust regression predicting z-scored regional GMV with age for each region from the Desikan atlas for all 4 runs and correlated the coefficients for age between RA and each of the RAC inputs.
Despite this bias, RAC estimates show similar developmental trajectories within this age range for overall tissue volumes (Fig. 2A) as well as for regional estimates of gray matter volume (Fig. 2B; min r2=0.78). We fit a robust regression predicting z-scored regional GMV with age for each region from the Desikan atlas for all 4 runs and correlated the coefficients for age between RA and each of the RAC inputs.
Conclusions:
Recon-all clinical is an exciting new tool, but the exclusion of child and adolescent data from the training set appears to result in biased estimates in children and adolescents. The underestimate of GMV matches the trajectory of neurodevelopment where regional gray matter volumes peak between 2 and 22 years of age (Bethlehem et al., 2022). Similarly, RAC with T1w inputs may overestimate WMV because WMV does not peak until adulthood. It is not clear why this relationship is inverted for RACT2. The relatively modest bias observed here would likely be larger in a sample with younger participants. Recon-all clinical may be usable with caution in child and adolescent samples, but a more robust solution would be to include data from children and adolescents in the training set.
Modeling and Analysis Methods:
Methods Development 1
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Cortical Anatomy and Brain Mapping 2
Keywords:
Development
Morphometrics
STRUCTURAL MRI
Other - Deep learning
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?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
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.
Gopinath, K., Greve, D.N., Magdamo, C., Arnold, S., Das, S. Puonti, O., Iglesias, J.E. (2024). “Recon-all-clinical”: Cortical surface reconstruction and analysis of heterogeneous clinical brain MRI. arXiv. https://doi.org/10.48550/arXiv.2409.03889