Parsing Disease Heterogeneity: GAN-based Normative Modeling for Personalized Insights
Poster No:
1504
Submission Type:
Abstract Submission
Authors:
Sai Spandana Chintapalli1, Sindhuja Govindarajan1, Haochang Shou2, Hao Huang3, Christos Davatzikos2
Institutions:
1Artificial Intelligence in Biomedical Imaging Laboratory, University of Pennsylvania, Philadelphia, PA, 2Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, 3Department of Radiology, Children’s Hospital of Philadelphia, Philadelphia, PA
First Author:
Sai Spandana Chintapalli
Artificial Intelligence in Biomedical Imaging Laboratory, University of Pennsylvania
Philadelphia, PA
Artificial Intelligence in Biomedical Imaging Laboratory, University of Pennsylvania
Philadelphia, PA
Co-Author(s):
Sindhuja Govindarajan
Artificial Intelligence in Biomedical Imaging Laboratory, University of Pennsylvania
Philadelphia, PA
Artificial Intelligence in Biomedical Imaging Laboratory, University of Pennsylvania
Philadelphia, PA
Introduction:
Neuroimaging studies have reported structural and functional brain heterogeneity in Alzheimer's disease (AD), traumatic brain injury (TBI), and other disorders1,2, that leads to diagnostic and prognostic uncertainty. Normative modeling, which computes individual-level deviations in brain measures from a reference sample to infer personalized disease effects, can help parse this heterogeneity3. Traditional univariate normative modeling methods like Gaussian process regression (GPR) ignore multivariate interactions between brain measures. While multivariate techniques like adversarial autoencoders (AAE) might have low specificity to disease effects as they are trained solely on the reference sample4. In both cases, the computed deviations might capture disease unrelated effects due to inter-individual brain differences. To overcome this, we propose a Generative Adversarial Network (GAN) based normative modeling framework (Fig.1.a) that synthesizes patient-specific controls by removing disease-related variations from patient's brain measures. Deviation of the patient from the synthesized disease-free control acts as an image-based biomarker that is sensitive to disease effects and severity.
Methods:
We adapt the pix2pix GAN5 to translate a patient (PT) to a corresponding healthy control (CN). Training requires paired data, i.e., neuroimaging-derived brain measures of individuals with and without disease. However, in reality, subjects either have the disease or do not. Hence, we synthetically simulate PT from a known reference sample of CN and use the pseudo-synthetic PT and real CN pairs for model training. To implement this method to study neuroanatomical heterogeneity, we select a reference population of 5600 CN from the ISTAGING consortium6 without pre-existing health conditions. Our neuroanatomical measures are the 139 region of interest (ROI) volumes (119 gray matter, 17 white matter) computed using a multi-atlas segmentation technique7. To simulate PT, for each CN, we randomly introduce 10-80% atrophy in a random set of ROIs while preserving clinical covariate effects. The model is optimized to translate pseudo-synthetic PT to real CN. During inference, the model synthesizes a CN for each real PT, and the difference between the two relates to real disease effects. For validation, we use held-out data and compare the estimated deviations between CN and subjects with different health conditions such as hypertension, diabetes, and TBI. Additionally, for performance assessment, we compare the deviations derived from the GAN model with those from GPR and AAE for AD classification. We select 100 CN and 100 AD participants from the OASIS8 dataset and compute their deviations across the 139 ROI volumes using GAN, GPR, and AAE models (pretrained on the ISTAGING dataset). Support vector machine is used to assess the overall discriminative power of the derived deviations in AD classification.
Results:
Fig. 1.b shows statistically significant differences in GAN-derived deviations for various health conditions compared to reference CN. Fig.2.a&b show that GAN-derived deviations are larger than those from other models and have stronger correlation with MMSE, brain age gap9, and SPARE-AD10. This shows that the GAN model captures disease-related abnormality in ROI volumes. Additionally, GAN-derived deviations improve AD classification (Fig.2.c). Finally, we note that the deviations derived by the GAN model in AD group reflect medial-temporal brain atrophy, a hallmark of AD (Fig.2.d).
Conclusions:
In conclusion, the GAN-based normative modeling technique is a useful tool for parsing individual-level brain heterogeneity. By leveraging self-supervised training with pseudo-synthetically simulated patient data, this method effectively detects disease-related effects in real-world conditions. This innovative approach holds promise for improving personalized pathology detection.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 2
Modeling and Analysis Methods:
Classification and Predictive Modeling
Methods Development 1
Multivariate Approaches
Keywords:
Computational Neuroscience
Data analysis
Degenerative Disease
Machine Learning
Other - Normative Modeling, Disease Heterogeneity, Generative Adversarial Networks (GANs)
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:
Provide references using APA citation style.
2) Segal, A., Parkes, L., Aquino, K., Kia, S. M., Wolfers, T., Franke, B., ... & Fornito, A. (2023). Regional, circuit and network heterogeneity of brain abnormalities in psychiatric disorders. Nature Neuroscience, 26(9), 1613-1629.
3) S. Rutherford, S. M. Kia, T. Wolfers, et al., “The normative modeling framework for compu-
tational psychiatry,” Nature protocols 17(7), 1711–1734 (2022).
4) Rutherford, S., Kia, S. M., Wolfers, T., Fraza, C., Zabihi, M., Dinga, R., ... & Marquand, A. F. (2022). The normative modeling framework for computational psychiatry. Nature protocols, 17(7), 1711-1734.
5) Isola, P., Zhu, J. Y., Zhou, T., & Efros, A. A. (2017). Image-to-image translation with conditional adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 1125-1134).
6) Habes, M., Pomponio, R., Shou, H., Doshi, J., Mamourian, E., Erus, G., ... & iSTAGING consortium, the Preclinical AD consortium, the ADNI, and the CARDIA studies. (2021). The Brain Chart of Aging: machine‐learning analytics reveals links between brain aging, white matter disease, amyloid burden, and cognition in the iSTAGING consortium of 10,216 harmonized MR scans. Alzheimer's & Dementia, 17(1), 89-102.
7) Doshi, J., Erus, G., Ou, Y., Resnick, S. M., Gur, R. C., Gur, R. E., ... & Alzheimer's Neuroimaging Initiative. (2016). MUSE: MUlti-atlas region Segmentation utilizing Ensembles of registration algorithms and parameters, and locally optimal atlas selection. Neuroimage, 127, 186-195.
8) LaMontagne, P. J., Benzinger, T. L., Morris, J. C., Keefe, S., Hornbeck, R., Xiong, C., ... & Marcus, D. (2019). OASIS-3: longitudinal neuroimaging, clinical, and cognitive dataset for normal aging and Alzheimer disease. medrxiv, 2019-12.
9) Guo, X., Ding, Y., Xu, W., Wang, D., Yu, H., Lin, Y., ... & Zhang, Y. (2024). Predicting brain age gap with radiomics and automl: A Promising approach for age-Related brain degeneration biomarkers. Journal of Neuroradiology, 51(3), 265-273.
10) Davatzikos, C., Xu, F., An, Y., Fan, Y., & Resnick, S. M. (2009). Longitudinal progression of Alzheimer's-like patterns of atrophy in normal older adults: the SPARE-AD index. Brain, 132(8), 2026-2035.