Interpretable Machine Learning for fMRI Classification in the Context of Movement Disorders
Poster No:
1121
Submission Type:
Abstract Submission
Authors:
Mariya Shumska1,2,3, Jelle Dalenberg1,3, Remco Renken4, Marina de Koning-Tijssen1,3, Michael Biehl2
Institutions:
1Expertise Centre Movement Disorders Groningen, University Medical Center Groningen, Groningen, Netherlands, 2Bernoulli Institute for Mathematics, Computer Science and AI, University of Groningen, Groningen, Netherlands, 3Department of Neurology, University Medical Center Groningen, Groningen, Netherlands, 4Department of Neuroscience, University Medical Center Groningen,, Groningen, Netherlands
First Author:
Mariya Shumska
Expertise Centre Movement Disorders Groningen, University Medical Center Groningen|Bernoulli Institute for Mathematics, Computer Science and AI, University of Groningen|Department of Neurology, University Medical Center Groningen
Groningen, Netherlands|Groningen, Netherlands|Groningen, Netherlands
Expertise Centre Movement Disorders Groningen, University Medical Center Groningen|Bernoulli Institute for Mathematics, Computer Science and AI, University of Groningen|Department of Neurology, University Medical Center Groningen
Groningen, Netherlands|Groningen, Netherlands|Groningen, Netherlands
Co-Author(s):
Jelle Dalenberg
Expertise Centre Movement Disorders Groningen, University Medical Center Groningen|Department of Neurology, University Medical Center Groningen
Groningen, Netherlands|Groningen, Netherlands
Expertise Centre Movement Disorders Groningen, University Medical Center Groningen|Department of Neurology, University Medical Center Groningen
Groningen, Netherlands|Groningen, Netherlands
Remco Renken
Department of Neuroscience, University Medical Center Groningen,
Groningen, Netherlands
Department of Neuroscience, University Medical Center Groningen,
Groningen, Netherlands
Marina de Koning-Tijssen
Expertise Centre Movement Disorders Groningen, University Medical Center Groningen|Department of Neurology, University Medical Center Groningen
Groningen, Netherlands|Groningen, Netherlands
Expertise Centre Movement Disorders Groningen, University Medical Center Groningen|Department of Neurology, University Medical Center Groningen
Groningen, Netherlands|Groningen, Netherlands
Michael Biehl
Bernoulli Institute for Mathematics, Computer Science and AI, University of Groningen
Groningen, Netherlands
Bernoulli Institute for Mathematics, Computer Science and AI, University of Groningen
Groningen, Netherlands
Introduction:
Hyperkinetic movement disorders are characterized by excessive, involuntary movements that significantly impair patients' quality of life. Among these, cortical myoclonus manifests as brief, shock-like jerks caused by abnormal electrical discharges in the cerebral cortex (Fahn, 1986; Latorre et al., 2018). Yet, the precise mechanisms underlying this activity remain unclear. Clinically, myoclonus is significant as its presence can guide the diagnosis of underlying conditions, including genetic disorders, encephalopathy, dementias, infectious diseases, drug-induced syndromes, metabolic abnormalities, and seizure disorders. This study aims to assist clinicians with a diagnosis of cortical myoclonus and contribute to the understanding of this disorder with the use of fMRI data and interpretable machine learning.
Methods:
We analyzed resting-state fMRI data collected by the Expertise Centre Movement Disorders Groningen (van der Stouwe et al., 2021; Dalenberg et al., 2024) from 16 patients with cortical myoclonus, whose diagnoses were confirmed by independent experts, and 16 healthy controls matched by sex and age. To summarize the data, we used 10 major resting-state networks identified in (Smith et al., 2009) through Independent Component Analysis, extracting 10 signals for each participant. Pairwise correlations between these signals were computed to create connectivity matrices, which were subsequently used to construct weighted graphs. From these graphs, we derived local and global features, including degree, clustering coefficient, betweenness centrality, and global efficiency (Wang, Zuo, & He, 2010).
To classify the data and examine differences between myoclonus patients and healthy controls, we employed the Generalized Matrix Learning Vector Quantization (GMLVQ) algorithm (Schneider, Biehl, & Hammer, 2009). This robust framework learns prototypes - representative examples for each class - by using an adaptive distance metric that evaluates the relevance of individual features and feature pair combinations for accurate classification. Furthermore, GMLVQ enhances interpretability by allowing the visualization of prototypes and feature relevances.
To classify the data and examine differences between myoclonus patients and healthy controls, we employed the Generalized Matrix Learning Vector Quantization (GMLVQ) algorithm (Schneider, Biehl, & Hammer, 2009). This robust framework learns prototypes - representative examples for each class - by using an adaptive distance metric that evaluates the relevance of individual features and feature pair combinations for accurate classification. Furthermore, GMLVQ enhances interpretability by allowing the visualization of prototypes and feature relevances.
·Summary of the methodology for resting-state fMRI analysis.
Results:
GMLVQ achieved 100% classification accuracy after leave-one-out cross-validation. Generally, we observed that the majority of resting-state networks of myoclonus patients tend to have lower degrees and clustering coefficients but higher betweenness centrality compared to those of healthy controls. Such a combination suggests that despite relatively sparse connections and the lack of local influence of certain networks, they remain structurally critical for maintaining global graph connectivity. Notably, in the healthy group, only the executive control network had this property.
The most important features distinguishing myoclonus patients from controls were associated with the sensorimotor network, including the supplementary motor area, sensorimotor cortex, and secondary somatosensory cortex. Specifically, the clustering coefficient and betweenness centrality were the most informative metrics. The examination of prototypes revealed that the former was lower in the myoclonus group while the latter was higher. This indicates that, in patients, the sensorimotor network is embedded within a less cohesive community of networks but plays a more critical role as a communication bridge between them.
The most important features distinguishing myoclonus patients from controls were associated with the sensorimotor network, including the supplementary motor area, sensorimotor cortex, and secondary somatosensory cortex. Specifically, the clustering coefficient and betweenness centrality were the most informative metrics. The examination of prototypes revealed that the former was lower in the myoclonus group while the latter was higher. This indicates that, in patients, the sensorimotor network is embedded within a less cohesive community of networks but plays a more critical role as a communication bridge between them.
·Left: average feature relevance and standard deviation across LOOCV folds. Right: average prototypes; standard deviation for all features was in range [0.03, 0.10] for both classes.
Conclusions:
This proof of concept shows that graph features derived from inter-network connectivity and the GMLVQ algorithm can be used to distinguish myoclonus patients from healthy controls. Most resting-state networks in patients showed lower degrees and clustering coefficients but higher betweenness centrality. The sensorimotor network in particular exhibited such properties and its features were assigned the greatest importance for correct classification. The proposed methodology establishes a foundation for future research into inter-network communication in other movement disorders with larger patient cohorts.
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
fMRI Connectivity and Network Modeling 2
Keywords:
FUNCTIONAL MRI
Machine Learning
Movement Disorder
Other - Interpretable Classification; Graph Theory
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.
2. Fahn, S. (1986). Definition and classification of myoclonus. Advances in neurology, 43 , 1–5.
3. Latorre, A., Rocchi, L., Berardelli, A., Rothwell, J. C., Bhatia, K. P., & Cordivari, C. (2018). Reappraisal of cortical myoclonus: A retrospective study of clinical neurophysiology. Movement Disorders, 33 (2), 339–341.
4. Schneider, P., Biehl, M., & Hammer, B. (2009, December). Adaptive relevance matrices in learning vector quantization. Neural Computation, 21 (12), 3532–3561.
5. Smith, S. M., Fox, P. T., Miller, K. L., Glahn, D. C., Fox, P. M., Mackay, C. E., . . . Beckmann, C. F. (2009). Correspondence of the brain’s functional architecture during activation and rest. Proceedings of the National Academy of Sciences, 106 (31), 13040-13045.
6. van der Stouwe, A. M., Tuitert, I., Giotis, I., Calon, J., Gannamani, R., Dalenberg, J. R., . . . Tijssen, M. A. (2021). Next move in movement disorders (nemo): developing a computer-aided classification tool for hyperkinetic movement disorders. BMJ open, 11 (10), e055068.
7. Wang, J., Zuo, X., & He, Y. (2010). Graph-based network analysis of resting-state functional MRI. Frontiers in systems neuroscience, 4, 16.