Machine Learning-Based Estimation of Heart Rate Variability from BOLD-fMRI in Aged Cohorts
Poster No:
1878
Submission Type:
Abstract Submission
Authors:
Abdoljalil Addeh1,2,3,4, Karen Ardila1,2,3,4, Rebecca J Williams5, G. Bruce Pike3,4,6, M. Ethan MacDonald1,2,3,4
Institutions:
1Department of Biomedical Engineering, Schulich School of Engineering, University of Calgary, Calgary, Canada, 2Department of Electrical & Software Engineering, Schulich School of Engineering, University of Calgary, Calgary, Canada, 3Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary, Calgary, Canada, 4Department of Radiology, Cumming School of Medicine, University of Calgary, Calgary, Canada, 5Brain-Behaviour Research Group, University of, New England, New South Wales, 6Department of Clinical Neurosciences, Cumming School of Medicine, University of Calgary, Calgary, Canada
First Author:
Abdoljalil Addeh
Department of Biomedical Engineering, Schulich School of Engineering, University of Calgary|Department of Electrical & Software Engineering, Schulich School of Engineering, University of Calgary|Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary|Department of Radiology, Cumming School of Medicine, University of Calgary
Calgary, Canada|Calgary, Canada|Calgary, Canada|Calgary, Canada
Department of Biomedical Engineering, Schulich School of Engineering, University of Calgary|Department of Electrical & Software Engineering, Schulich School of Engineering, University of Calgary|Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary|Department of Radiology, Cumming School of Medicine, University of Calgary
Calgary, Canada|Calgary, Canada|Calgary, Canada|Calgary, Canada
Co-Author(s):
Karen Ardila
Department of Biomedical Engineering, Schulich School of Engineering, University of Calgary|Department of Electrical & Software Engineering, Schulich School of Engineering, University of Calgary|Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary|Department of Radiology, Cumming School of Medicine, University of Calgary
Calgary, Canada|Calgary, Canada|Calgary, Canada|Calgary, Canada
Department of Biomedical Engineering, Schulich School of Engineering, University of Calgary|Department of Electrical & Software Engineering, Schulich School of Engineering, University of Calgary|Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary|Department of Radiology, Cumming School of Medicine, University of Calgary
Calgary, Canada|Calgary, Canada|Calgary, Canada|Calgary, Canada
G. Bruce Pike
Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary|Department of Radiology, Cumming School of Medicine, University of Calgary|Department of Clinical Neurosciences, Cumming School of Medicine, University of Calgary
Calgary, Canada|Calgary, Canada|Calgary, Canada
Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary|Department of Radiology, Cumming School of Medicine, University of Calgary|Department of Clinical Neurosciences, Cumming School of Medicine, University of Calgary
Calgary, Canada|Calgary, Canada|Calgary, Canada
M. Ethan MacDonald
Department of Biomedical Engineering, Schulich School of Engineering, University of Calgary|Department of Electrical & Software Engineering, Schulich School of Engineering, University of Calgary|Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary|Department of Radiology, Cumming School of Medicine, University of Calgary
Calgary, Canada|Calgary, Canada|Calgary, Canada|Calgary, Canada
Department of Biomedical Engineering, Schulich School of Engineering, University of Calgary|Department of Electrical & Software Engineering, Schulich School of Engineering, University of Calgary|Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary|Department of Radiology, Cumming School of Medicine, University of Calgary
Calgary, Canada|Calgary, Canada|Calgary, Canada|Calgary, Canada
Introduction:
Recent advancements in machine learning (ML) have introduced novel approaches that leverage resting-state fMRI signals to directly estimate Heart Rate Variation (HRV) waveforms, particularly in young adult populations (Bayrak et al., 2021), thus reducing reliance on external physiological measurements. Despite these developments, HRV estimation in aged populations remains unexplored (Bayrak et al., 2021).
This study presents a first attempt to estimate HRV from resting-state fMRI data in older adults, a particularly challenging cohort in neuroimaging studies due to age-related structural and functional brain changes (MacDonald & Pike, 2021). We propose an ML framework that utilizes dynamic functional connectivity (dFC)-based brain atlases to better capture the intricate temporal dynamics of HRV by integrating information from both gray and white matter. This approach is expected to significantly improve HRV estimation accuracy, with performance metrics surpassing those of static connectivity-based atlases used in previous HRV estimation studies by at least 5%.
This study presents a first attempt to estimate HRV from resting-state fMRI data in older adults, a particularly challenging cohort in neuroimaging studies due to age-related structural and functional brain changes (MacDonald & Pike, 2021). We propose an ML framework that utilizes dynamic functional connectivity (dFC)-based brain atlases to better capture the intricate temporal dynamics of HRV by integrating information from both gray and white matter. This approach is expected to significantly improve HRV estimation accuracy, with performance metrics surpassing those of static connectivity-based atlases used in previous HRV estimation studies by at least 5%.
Methods:
This study utilized a novel approach for HRV reconstruction, employing a hybrid model combining one-dimensional Convolutional Neural Networks (1D-CNN) and Gated Recurrent Units (GRU). The model analyzed blood oxygen level-dependent (BOLD) signals from 628 ROIs, comprising 518 cortical, 62 subcortical, and 48 white matter ROIs, derived from dynamic functional connectivity (Peng et al., 2023) and diffusion tensor imaging-based atlases (Hua et al., 2008). A 65-timepoint (TR) sliding window technique was implemented to estimate HRV at the 10th point within the window, optimizing the model's ability to capture HRV-related BOLD fluctuations. The input data and windowing methodology mirrored our previous Respiratory Variation reconstruction approach (Addeh et al., 2023; Addeh, Vega, Morshedi, et al., 2024).
This study used 700 resting-state fMRI scans from 429 HCP-A participants, selected for high-quality cardiac signals for training and evaluation (Addeh, Vega, Ardila, et al., 2024). We performed ten-fold cross-validation, using nine folds for training and one for testing, repeated across all folds. Model performance was averaged across iterations using MAE, MSE, Pearson correlation, and Dynamic Time Warping (DTW) metrics to assess overall efficacy.
We evaluated the impact of ROI selection by incorporating 580 dynamic functional ROIs (Peng et al., 2023), 48 white matter ROIs (Hua et al., 2008), the Glasser atlas (360 static functional ROIs) (Glasser et al., 2016), and the Harvard-Oxford atlas (69 anatomical ROIs) (Makris et al., 2006). Comparisons of each model's performance based on specific ROI inputs were statistically validated using the Friedman test.
This study used 700 resting-state fMRI scans from 429 HCP-A participants, selected for high-quality cardiac signals for training and evaluation (Addeh, Vega, Ardila, et al., 2024). We performed ten-fold cross-validation, using nine folds for training and one for testing, repeated across all folds. Model performance was averaged across iterations using MAE, MSE, Pearson correlation, and Dynamic Time Warping (DTW) metrics to assess overall efficacy.
We evaluated the impact of ROI selection by incorporating 580 dynamic functional ROIs (Peng et al., 2023), 48 white matter ROIs (Hua et al., 2008), the Glasser atlas (360 static functional ROIs) (Glasser et al., 2016), and the Harvard-Oxford atlas (69 anatomical ROIs) (Makris et al., 2006). Comparisons of each model's performance based on specific ROI inputs were statistically validated using the Friedman test.
Results:
The 1D-CNN + GRU network's accuracy in reconstructing HRV time-series is shown in Figure 1, demonstrating high precision, especially during significant HRV fluctuations.
Figure 2 evaluates the model's performance across different configurations. Configurations integrating dynamic and white matter ROIs outperformed others, confirming the benefits of this approach for HRV estimation in BOLD-fMRI studies. Statistical analysis revealed significant differences between the Dynamic Functional + White Matter ROIs and Dynamic Functional ROIs Only (p-value < 0.001), and between the Dynamic Functional + White Matter ROIs and the Static Functional + White Matter ROIs (p-value < 0.01).
Figure 2 evaluates the model's performance across different configurations. Configurations integrating dynamic and white matter ROIs outperformed others, confirming the benefits of this approach for HRV estimation in BOLD-fMRI studies. Statistical analysis revealed significant differences between the Dynamic Functional + White Matter ROIs and Dynamic Functional ROIs Only (p-value < 0.001), and between the Dynamic Functional + White Matter ROIs and the Static Functional + White Matter ROIs (p-value < 0.01).
·Comparison of measured and reconstructed HRV waveforms using the proposed method across three cases (a, b, c). Performance metrics highlight the model’s accuracy in capturing HRV dynamics
·Violin plots illustrating the model's performance in reconstructing HRV across input configurations. The combination of Dynamic Functional ROIs with White Matter shows the best performance
Conclusions:
The results demonstrate that dFC-based brain atlases outperform static and anatomical ROIs by better capturing HRV's temporal dynamics. Adding white matter regions further improved accuracy, confirming their role in physiological confound correction.
Our proposed framework achieved over a 7% improvement in performance metrics, exceeding our 5% hypothesis. Future work will focus on refining network architecture, optimizing ROI selection, and validating on larger datasets to improve generalizability, particularly in aging populations.
Our proposed framework achieved over a 7% improvement in performance metrics, exceeding our 5% hypothesis. Future work will focus on refining network architecture, optimizing ROI selection, and validating on larger datasets to improve generalizability, particularly in aging populations.
Lifespan Development:
Aging 2
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling
Novel Imaging Acquisition Methods:
BOLD fMRI 1
Keywords:
Aging
fMRI CONTRAST MECHANISMS
Machine 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?
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.
Addeh, A., Vega, F., Medi, P. R., Williams, R. J., Pike, G. B., & MacDonald, M. E. (2023). Direct machine learning reconstruction of respiratory variation waveforms from resting state fMRI data in a pediatric population. NeuroImage, 269, 119904. https://doi.org/https://doi.org/10.1016/j.neuroimage.2023.119904
Addeh, A., Vega, F., Morshedi, A., Williams, R. J., Pike, G. B., & MacDonald, M. E. (2024). Machine learning‐based estimation of respiratory fluctuations in a healthy adult population using resting state BOLD fMRI and head motion parameters. Magnetic Resonance in Medicine, 1-15.
Bayrak, R. G., Hansen, C. B., Salas, J. A., Ahmed, N., Lyu, I., Huo, Y., & Chang, C. (2021). From Brain to Body: Learning Low-Frequency Respiration and Cardiac Signals from fMRI Dynamics. Medical Image Computing and Computer Assisted Intervention – MICCAI 2021, Cham.
Glasser, M. F., Coalson, T. S., Robinson, E. C., Hacker, C. D., Harwell, J., Yacoub, E., Ugurbil, K., Andersson, J., Beckmann, C. F., & Jenkinson, M. (2016). A multi-modal parcellation of human cerebral cortex. Nature, 536(7615), 171-178.
Hua, K., Zhang, J., Wakana, S., Jiang, H., Li, X., Reich, D. S., Calabresi, P. A., Pekar, J. J., van Zijl, P. C., & Mori, S. (2008). Tract probability maps in stereotaxic spaces: analyses of white matter anatomy and tract-specific quantification. Neuroimage, 39(1), 336-347.
MacDonald, M. E., & Pike, G. B. (2021). MRI of healthy brain aging: A review. NMR in Biomedicine, 34(9), e4564.
Makris, N., Goldstein, J. M., Kennedy, D., Hodge, S. M., Caviness, V. S., Faraone, S. V., Tsuang, M. T., & Seidman, L. J. (2006). Decreased volume of left and total anterior insular lobule in schizophrenia. Schizophrenia research, 83(2-3), 155-171.
Peng, L., Luo, Z., Zeng, L.-L., Hou, C., Shen, H., Zhou, Z., & Hu, D. (2023). Parcellating the human brain using resting-state dynamic functional connectivity. Cerebral Cortex, 33(7), 3575-3590.