A Novel Transformer Model for ADHD Subtyping and Brain Functional Groups Identification using fMRI
Poster No:
1399
Submission Type:
Abstract Submission
Authors:
Yanting Yang1, Beidi Zhao1, Nicha Dvornek2, Zhuohao Ni1, Varsha Sreenivasan1, Martin McKeown1, Yize Zhao3, Xiaoxiao Li1
Institutions:
1University of British Columbia, Vancouver, British Columbia, 2Yale University, New Haven, CT, 3School of Public Health, Yale University, New Haven, CT
First Author:
Co-Author(s):
Introduction:
Attention Deficit Hyperactivity Disorder (ADHD) is a heterogeneous neurodevelopmental disorder that varies between individuals, leading to distinct subtypes. ADHD subtyping is critical for tailoring clinical interventions and improving patient outcomes, as each subtype exhibits unique behavioral and neurobiological profiles. These subtypes are associated with differences in brain function networks/groups (FGs), reflecting distinct neural connectivity patterns (Saad, 2017). However, current data-driven analyses fail to identify subtype-specific FGs that underlie these differences. Addressing this gap is vital for developing precision medicine approaches that can inform personalized diagnostic and therapeutic strategies.
Methods:
Our work introduces a novel transformer-based deep learning architecture to distinguish the brain FGs associated with different ADHD subtypes. We term this method FGTF, which was applied to the ADHD-200 dataset, including fMRI data from 785 healthy controls and 460 individuals diagnosed with ADHD-Combined (ADHD-C) or ADHD-Inattentive (ADHD-I) subtypes. Data preprocessing involved standard parcellation techniques to generate functional connectivity matrices, which served as input for the model (Fig 1a). The model uses a transformer-based architecture incorporating adaptive clustering to dynamically identify brain FGs (Fig 1b). It includes learnable clustering tokens and a deep embedding clustering block (Fig 1c) to optimize the identification of brain region clusters. The model directly connects brain organization and classification outcomes, supervised by the cross-entropy loss. It dynamically assigns ROIs to clusters while maintaining flexibility to reflect subtype-specific variations, offering both personalized insights and group-level interpretability (Fig 1d).
Results:
Comparison studies (Fig 1e) demonstrate FGTF's exceptional performance in ADHD subtyping. It outperformed baseline methods in accuracy, F1 score, and sensitivity, which achieved statistically significant improvements. Chord diagrams (Fig 2a) show that the intra-cluster connections consistently exhibit higher co-association probabilities compared to inter-cluster connections. This highlights the model's ability to capture cohesive FGs within clusters, reinforcing their biological plausibility. Meta-analyses (Fig 2b) revealed that the identified FGs correspond to known cognitive and behavioral domains relevant to ADHD and that the terms identified vary between subtypes. For example, ADHD-C has a strong association with perception and motion while ADHD-I exhibits prominent term frequencies in speech, language, and pain. These findings align with established literature on the neural underpinnings of ADHD (Ptacek, 2019; Dahan, 2017; Kim, 2020; Stray, 2013), further supporting the biological relevance of the clusters. The predictive power of each functional group (Fig 2c) provides insight into the relative contribution of different FGs. The predictive contributions show variability between subtypes, with cluster 8 having the highest contribution in ADHD-C and cluster 3 having the highest contribution in ADHD-I.
Conclusions:
FGTF represents a transformative step in ADHD research, addressing the challenges of ADHD subtype classification by capturing their variability in brain FGs. Its ability to adaptively cluster brain regions uncovers biologically plausible FGs and offers a novel framework for understanding ADHD subtypes. By revealing distinct neurofunctional patterns, it provides critical insights into the neurobiological heterogeneity of ADHD. The model's superior classification performance and biologically meaningful interpretations highlight its potential for identifying subtype-specific biomarkers, which could drive the development of targeted interventions and personalized treatment strategies. This approach not only advances precision medicine for ADHD but also establishes a scalable paradigm for studying other neurological and cognitive disorders.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 2
fMRI Connectivity and Network Modeling 1
Keywords:
Attention Deficit Disorder
FUNCTIONAL MRI
Meta- Analysis
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:
Provide references using APA citation style.
2. Dahan, A. (2017). Evidence for deficient motor planning in ADHD. Scientific reports, 7(1), 9631.
3. Kan, X. (2022a). Fbnetgen: Task-aware gnn-based fmri analysis via functional brain network generation. In International Conference on Medical Imaging with Deep Learning (pp. 618-637). PMLR.
4. Kan, X. (2022b). Brain network transformer. Advances in Neural Information Processing Systems, 35, 25586-25599.
5. Kawahara, J. (2017). BrainNetCNN: Convolutional neural networks for brain networks; towards predicting neurodevelopment. NeuroImage, 146, 1038-1049.
6. Kim, O. H. (2000). Language characteristics of children with ADHD. Communication disorders quarterly, 21(3), 154-165.
7. Ptacek, R. (2019). Clinical implications of the perception of time in attention deficit hyperactivity disorder (ADHD): A review. Medical science monitor: international medical journal of experimental and clinical research, 25, 3918.
8. Saad, J. F. (2017). Regional brain network organization distinguishes the combined and inattentive subtypes of attention deficit hyperactivity disorder. NeuroImage: Clinical, 15, 383-390.
9. Stray, L. L. (2013). Motor regulation problems and pain in adults diagnosed with ADHD. Behavioral and Brain Functions, 9, 1-10.