A Foundation Model for Classification and Symptom Prediction in Childhood ADHD
Poster No:
263
Submission Type:
Abstract Submission
Authors:
Harinarayana Mellacheruvu1, Srikanth Ryali1, Yuan Zhang1, Vinod Menon1, Weidong Cai1
Institutions:
1Stanford University School of Medicine, Stanford, CA
First Author:
Co-Author(s):
Introduction:
Childhood Attention Deficit Hyperactivity Disorder (ADHD) is the most common neurodevelopmental disorder, with prevalence rates rising significantly in recent years (Cortese, 2023; Xu, 2018). Since ADHD is typically diagnosed through subjective self-reports, which are prone to observational biases, there has been a long-standing need for objective and quantitative biomarker-driven approaches to diagnosis (Gascon, 2022). While ADHD is modeled as a disorder involving disturbances in large-scale brain networks, with altered functional connectivity linked to its symptoms (Castellanos, 2012; Menon, 2011; Posner, 2014), neuroimaging-based classification models of ADHD have not yet achieved good performance, likely due to challenges with small sample sizes and data heterogeneity (Acuna, 2012; Cortese, 2021). To address these limitations, we propose to develop a robust foundation model (FM) for differentiating children with ADHD and typically developing (TD) children, leveraging recent advances in artificial intelligence and multiple large-scale neuroimaging datasets.
Methods:
Resting-state fMRI data was obtained from three large-scale datasets, including ADHD-200 (200 ADHD, 200 TD, 7-18 years old), ABCD (270 ADHD, 272 TD, 9-11 years old) and CMI-HBN (468 ADHD, 117 TD, 7-18 years old).
FM is trained through self-supervision on large unlabeled fMRI data, allowing it to learn complex latent spatiotemporal dynamics of brain activity. FM consists of three 1D convolution blocks, a temporal averaging operation that reduces the number of model parameters, and a sigmoid layer for classification. The model was pretrained on two large unlabeled datasets containing healthy participants, HCP (N=1075) and NKI-RS (N=982). We then fine-tuned FM on the ADHD-200, ABCD and CMI-HBN datasets to identify ADHD presence and symptom severity.
We also applied integrated gradients, an explainable AI (XAI) method, to FM, allowing us to determine importance of each brain region based on its contribution to ADHD prediction.
FM is trained through self-supervision on large unlabeled fMRI data, allowing it to learn complex latent spatiotemporal dynamics of brain activity. FM consists of three 1D convolution blocks, a temporal averaging operation that reduces the number of model parameters, and a sigmoid layer for classification. The model was pretrained on two large unlabeled datasets containing healthy participants, HCP (N=1075) and NKI-RS (N=982). We then fine-tuned FM on the ADHD-200, ABCD and CMI-HBN datasets to identify ADHD presence and symptom severity.
We also applied integrated gradients, an explainable AI (XAI) method, to FM, allowing us to determine importance of each brain region based on its contribution to ADHD prediction.
Results:
Classification of ADHD versus TD children: FM achieved good performance for ADHD classification in each dataset: cross validation accuracy of 74.9% and F1-score of 73.4% in the ADHD-200 cohort, accuracy of 64.0% and F1-score of 63.6% in the ABCD cohort, and accuracy of 64.1% and F1-score of 62.7% in the CMI-HBN cohort. Importantly, FM has significantly better performance than conventional machine learning models.
Brain features contributing to ADHD classification: XAI analysis identified the ventromedial and lateral prefrontal cortex, middle/inferior temporal gyrus, thalamus and visual cortex as within the top 20% of features underlying ADHD prediction in the ADHD-200 cohort (Figure 1). In the ABCD cohort, top features include the lateral prefrontal cortex, inferior temporal gyrus, postcentral gyrus and precuneus. In the CMI-HBN cohort, top features include the ventromedial prefrontal cortex, inferior temporal gyrus and visual cortex. Moreover, the feature importance rankings were significantly correlated across the three independent cohorts (all rs>0.2, ps<0.001), indicating that similar features contribute to distinguishing ADHD from TD across datasets.
ADHD symptom prediction: FM can significantly predict ADHD symptoms in each dataset (ADHD-200 Hyperactivity: r=0.29, p=1.3e-8; ABCD CBCL ADHD: r=0.25, p=5.9e-9, CMI-HBN Hyperactivity: r=0.154, p=9.8e-05).
Brain features contributing to ADHD classification: XAI analysis identified the ventromedial and lateral prefrontal cortex, middle/inferior temporal gyrus, thalamus and visual cortex as within the top 20% of features underlying ADHD prediction in the ADHD-200 cohort (Figure 1). In the ABCD cohort, top features include the lateral prefrontal cortex, inferior temporal gyrus, postcentral gyrus and precuneus. In the CMI-HBN cohort, top features include the ventromedial prefrontal cortex, inferior temporal gyrus and visual cortex. Moreover, the feature importance rankings were significantly correlated across the three independent cohorts (all rs>0.2, ps<0.001), indicating that similar features contribute to distinguishing ADHD from TD across datasets.
ADHD symptom prediction: FM can significantly predict ADHD symptoms in each dataset (ADHD-200 Hyperactivity: r=0.29, p=1.3e-8; ABCD CBCL ADHD: r=0.25, p=5.9e-9, CMI-HBN Hyperactivity: r=0.154, p=9.8e-05).
Conclusions:
Our findings demonstrate that FM can effectively distinguish children with ADHD from TD children and identify functional brain signatures. Notably, it outperformed both traditional machine learning models. This impressive performance was replicated across three independent large-scale multi-site neuroimaging datasets, underscoring the FM's robustness, reproducibility and generalizability in ADHD classification. Additionally, the FM can be efficiently fine-tuned to predict ADHD symptom severity, indicating its potentials to evolve into a transdiagnostic framework.
Disorders of the Nervous System:
Neurodevelopmental/ Early Life (eg. ADHD, autism) 1
Higher Cognitive Functions:
Executive Function, Cognitive Control and Decision Making
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Keywords:
Attention Deficit Disorder
Cognition
FUNCTIONAL MRI
Pediatric Disorders
Psychiatric Disorders
Other - deep learning, impulsivity, inattention, resting-state fMRI, prefrontal cortex
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.
Castellanos, F. X., & Proal, E. (2012). Large-scale brain systems in ADHD: beyond the prefrontal-striatal model. Trends Cogn Sci, 16(1), 17-26. doi:10.1016/j.tics.2011.11.007
Cortese, S., Aoki, Y. Y., Itahashi, T., Castellanos, F. X., & Eickhoff, S. B. (2021). Systematic Review and Meta-analysis: Resting-State Functional Magnetic Resonance Imaging Studies of Attention-Deficit/Hyperactivity Disorder. Journal of the American Academy of Child and Adolescent Psychiatry, 60(1), 61-75. doi:10.1016/j.jaac.2020.08.014
Cortese, S., Song, M. J., Farhat, L. C., Yon, D. K., Lee, S. W., Kim, M. S., . . . Solmi, M. (2023). Incidence, prevalence, and global burden of ADHD from 1990 to 2019 across 204 countries: data, with critical re-analysis, from the Global Burden of Disease study. Molecular Psychiatry, 28(11), 4823-4830. doi:10.1038/s41380-023-02228-3
Gascon, A., Gamache, D., St-Laurent, D., & Stipanicic, A. (2022). Do we over-diagnose ADHD in North America? A critical review and clinical recommendations. Journal of Clinical Psychology, 78(12), 2363-2380. doi:10.1002/jclp.23348
Menon, V. (2011). Large-scale brain networks and psychopathology: a unifying triple network model. Trends Cogn Sci, 15(10), 483-506. doi:10.1016/j.tics.2011.08.003
Posner, J., Park, C., & Wang, Z. S. (2014). Connecting the Dots: A Review of Resting Connectivity MRI Studies in Attention-Deficit/Hyperactivity Disorder. Neuropsychology Review, 24(1), 3-15. doi:10.1007/s11065-014-9251-z
Xu, G. F., Strathearn, L., Liu, B. Y., Yang, B. R., & Bao, W. (2018). Twenty-Year Trends in Diagnosed Attention-Deficit/Hyperactivity Disorder Among US Children and Adolescents, 1997-2016. Jama Network Open, 1(4). doi:ARTN e18147110.1001/jamanetworkopen.2018.1471