It’s not a bug, it’s a feature: Estimating bone mineral density from T1-weighted MR images
Poster No:
2101
Submission Type:
Abstract Submission
Authors:
Polona Kalc1, Felix Hoffstaedter2, Eileen Luders3, Christian Gaser4, Robert Dahnke4
Institutions:
1University Hospital Jena, Jena, Germany, 2Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich, Jülich, NRW, 3University of Auckland, Auckland, New Zealand, 4Jena University Hospital, Jena, Germany
First Author:
Co-Author(s):
Felix Hoffstaedter
Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Jülich, NRW
Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Jülich, NRW
Introduction:
No brain is an island and, as such, it is intricately connected to other tissues and organs [1], yet, the neuroimaging community largely dismisses the information from the closest bony structure surrounding the brain. More specifically, the skull may contain markers for bone mineral density and bone metabolism which, in turn, may affect emotion and cognition [2–3], and even the risk of Alzheimer's disease [4–6]. However, the bone mineral density (hereafter referred to as BMD) measures are typically not available in open-source brain imaging databases. Here we present a novel approach to calculate a proxy for bone mineral density from the skull based on a single T1-weighted image of the human brain.
Methods:
Sample
This research has been conducted using a subsample of 1,000 healthy subjects (M = 63.81 ± 6.27 years, age range: 46–79, 50% women) from the UK Biobank (UKB application 41655). Imaging parameters are provided in [7]. The sample was used to extract and validate our BMD estimate against the head BMD measure, obtained by dual energy X-ray absorptiometry (DXA) [8]. In addition, a subsample of 63 subjects from OASIS-3 (M = 69.27 ± 7.32 years, age range: 48–85, 54% women) was used to determine the retest reliability. Imaging parameters are provided in [9].
Data processing
All T1-weighted brain images were processed using SPM12 (https://www.fil.ion.ucl.ac.uk/spm/) and Matlab 2021a (https://www.mathworks.com), which produced the following segments: gray matter, white matter, cerebrospinal fluid, skull, soft head tissue, and background. Of note, instead of the default setting (3 mm), we set SPM's 'samp' parameter to 5 mm to ensure that non-brain tissues were properly classified. Misclassified skull segments were corrected by morphological operations.
We created a template atlas of the skull regions by averaging affine and intensity-normalized T1-weighted images as well as CT data of OASIS-3, and manually labeling the skull segment tissue probability map using Slicer3D [10]. The atlas was mapped into individual space using the linear transformation from the SPM segmentation.
The individual BMD measures were derived from the corrected SPM skull segment, by quantifying the mean intensity across the entire segment labeled as skull (or within a specific skull region), normalized by the median intensity of the CSF segment.
This research has been conducted using a subsample of 1,000 healthy subjects (M = 63.81 ± 6.27 years, age range: 46–79, 50% women) from the UK Biobank (UKB application 41655). Imaging parameters are provided in [7]. The sample was used to extract and validate our BMD estimate against the head BMD measure, obtained by dual energy X-ray absorptiometry (DXA) [8]. In addition, a subsample of 63 subjects from OASIS-3 (M = 69.27 ± 7.32 years, age range: 48–85, 54% women) was used to determine the retest reliability. Imaging parameters are provided in [9].
Data processing
All T1-weighted brain images were processed using SPM12 (https://www.fil.ion.ucl.ac.uk/spm/) and Matlab 2021a (https://www.mathworks.com), which produced the following segments: gray matter, white matter, cerebrospinal fluid, skull, soft head tissue, and background. Of note, instead of the default setting (3 mm), we set SPM's 'samp' parameter to 5 mm to ensure that non-brain tissues were properly classified. Misclassified skull segments were corrected by morphological operations.
We created a template atlas of the skull regions by averaging affine and intensity-normalized T1-weighted images as well as CT data of OASIS-3, and manually labeling the skull segment tissue probability map using Slicer3D [10]. The atlas was mapped into individual space using the linear transformation from the SPM segmentation.
The individual BMD measures were derived from the corrected SPM skull segment, by quantifying the mean intensity across the entire segment labeled as skull (or within a specific skull region), normalized by the median intensity of the CSF segment.
Results:
The obtained BMD estimate of the skull was validated against the BMD measure of the head (UKB data-field 23226-2.0), left femoral neck (UKB data-field 23299-2.0), as well as the total body BMD (UKB data-field 23239-2.0). Our proxy skull BMD estimate is highly correlated to the head (r =.71, p<.001), as well as total BMD (r =.60, p<.001), and is moderately associated with BMD of the left femoral neck (r =.41, p<.001). The results of the validation are shown in Figure 2.
In addition, we calculated the retest reliability of the BMD estimate of the skull using T1-weighted images from the OASIS-3 dataset acquired at two time points within an interval of less than 3 months (r = 0.97; p <.001).
In addition, we calculated the retest reliability of the BMD estimate of the skull using T1-weighted images from the OASIS-3 dataset acquired at two time points within an interval of less than 3 months (r = 0.97; p <.001).
Conclusions:
We developed an approximation of skull BMD by making use of tissue classes that are normally discarded when processing brain MR images. The estimation of skull BMD from T1-weighted brain images may serve as a proxy for a person's total BMD in research studies where such information would be relevant but has not been collected. Moreover, the extracted measure could provide valuable insights into the interconnectedness of bones and brain.
Modeling and Analysis Methods:
Classification and Predictive Modeling
Image Registration and Computational Anatomy 2
Methods Development
Segmentation and Parcellation
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Anatomy and Functional Systems 1
Keywords:
Morphometrics
STRUCTURAL MRI
Other - bone mineral density, skull
1|2Indicates the priority used for review
Provide references using author date format
[2] Nakamura, M., Imaoka, M., & Takeda, M. (2021), ‘Interaction of bone and brain: Osteocalcin and cognition’, International Journal of Neuroscience, vol. 131, no. 11, pp. 1115–1123.
[3] Obri, A., Khrimian, L., Karsenty, G., & Oury, F. (2018), ‘Osteocalcin in the brain: From embryonic development to age-related decline in cognition’, Nature Reviews Endocrinology, vol. 14, no. 3, pp. 174–182.
[4] Kostev, K., Hadji, P., & Jacob, L. (2018), ‘Impact of Osteoporosis on the Risk of Dementia in Almost 60,000 Patients Followed in General Practices in Germany’, Journal of Alzheimer’s Disease, vol. 65, no. 2, pp. 401–407.
[5] Xiao, T., Ghatan, S., Mooldijk, S. S., Trajanoska, K., Oei, L., Gomez, M. M., Ikram, M. K., Rivadeneira, F., & Ikram, M. A. (2023), ‘Association of Bone Mineral Density and Dementia: The Rotterdam Study. Neurology’, vol. 100, no. 20, e2125–e2133.
[6] Zhang, X., Hsu, W. W. Q., Sing, C.-W., Li, G. H. Y., Tan, K. C. B., Kung, A. W. C., Wong, J. S. H., Wong, I. C.-K., & Cheung, C.-L. (2022), ‘Low Bone Mineral Density With Risk of Dementia: A Prospective Cohort Study’, Journal of the American Medical Directors Association, vol. 23, no. 10, 1719.e9-1719.e19.
[7] Smith, S.M., Alfaro-Almagro, F., & Miller, K.L. (2020), ‘UK Biobank imaging documentation, Version 1.8.’, biobank.ctsu.ox.ac.uk/crystal/docs/brain_mri.pdf
[8] UK Biobank (2015), Imaging modality DXA. Retrieved from https://biobank.ndph.ox.ac.uk/showcase/ukb/docs/DXA_explan_doc.pdf.
[9] LaMontagne, P. J., Benzinger, T. Ls., Morris, J. C., Keefe, S., Hornbeck, R., Xiong, C., Grant, E., Hassenstab, J., Moulder, K., Vlassenko, A. G., Raichle, M. E., Cruchaga, C., & Marcus, D. (2019), ‘OASIS-3: Longitudinal Neuroimaging, Clinical, and Cognitive Dataset for Normal Aging and Alzheimer Disease’, [Preprint]. Radiology and Imaging.
[10] Fedorov, A., Beichel, R., Kalpathy-Cramer, J., Finet, J., Fillion-Robin, J.-C., Pujol, S., Bauer, C., Jennings, D., Fennessy, F., Sonka, M., Buatti, J., Aylward, S., Miller, J. V., Pieper, S., & Kikinis, R. (2012), ‘3D Slicer as an image computing platform for the Quantitative Imaging Network’, Magnetic Resonance Imaging, vol. 30, no. 9, pp. 1323–1341.