Comparing MedSigLIP and Structured Connectivity Models for ADHD and Schizophrenia Classification

• Eronides F. da Silva Neto UFPE
• Beatriz L. Bonafini CESAR School
• Breno C. Bispo UFPE
• Juliano B. Lima UFPE

DOI: https://doi.org/10.5753/sbcas.2026.21443

Resumo

The diagnosis of ADHD and schizophrenia is evolving with the integration of functional magnetic resonance imaging and machine learning. This study compares structured connectivity models (graph and hypergraph) with MedSigLIP foundation-model embeddings for disorder classification. Subject-level representations were extracted from the ADHD-200 and COBRE datasets and evaluated using 5-fold cross-validation. Among four classifiers, SVM consistently achieved the best performance. For ADHD, SVM reached an mAUC of 64.89%, approaching a graph-based baseline. For schizophrenia, it achieved an mAUC of 62.48%. These results indicate a competitive alternative to handcrafted connectivity features, even without disorder-specific pretraining.

Referências

Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F. L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., et al. (2023). Gpt-4 technical report. arXiv preprint arXiv:2303.08774.

Al-Wardat, M., Etoom, M., Almhdawi, K. A., Hawamdeh, Z., and Khader, Y. (2024). Prevalence of attention-deficit hyperactivity disorder in children, adolescents and adults in the middle east and north africa region: a systematic review and metaanalysis. BMJ open, 14(1):e078849.

Anwar, A., Mustafa, A. M., Abdou, K., A. Rabie, M., El-Shiekh, R. A., and El-Dessouki, A. M. (2025). A comprehensive review on schizophrenia: epidemiology, pathogenesis, diagnosis, conventional treatments, and proposed natural compounds used for management. Naunyn-Schmiedeberg’s Archives of Pharmacology, pages 1–25.

Bassett, D. S. and Sporns, O. (2017). Network neuroscience. Nature neuroscience, 20(3):353–364.

Du, Y., Fang, S., He, X., and Calhoun, V. D. (2024). A survey of brain functional network extraction methods using fmri data. Trends in Neurosciences, 47(8):608–621.

Han, X., Xue, R., Feng, J., Feng, Y., Du, S., Shi, J., and Gao, Y. (2025). Hypergraph foundation model for brain disease diagnosis. IEEE Transactions on Neural Networks and Learning Systems, 36(10):17702–17716.

Ji, J., Ren, Y., and Lei, M. (2022). Fc–hat: Hypergraph attention network for functional brain network classification. Information Sciences, 608:1301–1316.

Liu, B., Zhan, L.-M., Xu, L., Ma, L., Yang, Y., and Wu, X.-M. (2021). Slake: A semantically-labeled knowledge-enhanced dataset for medical visual question answering. In 2021 IEEE 18th international symposium on biomedical imaging (ISBI), pages 1650–1654. IEEE.

M. Milham, D. Fair, M. Mennes, S. Mostofsky (2012). The adhd-200 consortium: a model to advance the translational potential of neuroimaging in clinical neuroscience. Frontiers in systems neuroscience, 6:62.

Noah, A. A. and Sedky, H. E. (2025). New frontiers in pharmacological treatment of attention-deficit hyperactivity disorder. Naunyn-Schmiedeberg’s Archives of Pharmacology, pages 1–11.

Sellergren, A., Kazemzadeh, S., Jaroensri, T., Kiraly, A., Traverse, M., Kohlberger, T., Xu, S., Jamil, F., Hughes, C., Lau, C., et al. (2025). Medgemma technical report. arXiv preprint arXiv:2507.05201.

Xiao, L., Wang, J., Kassani, P. H., Zhang, Y., Bai, Y., Stephen, J. M., Wilson, T. W., Calhoun, V. D., and Wang, Y.-P. (2020). Multi-hypergraph learning-based brain functional connectivity analysis in fmri data. IEEE Transactions on Medical Imaging, 39(5):1746–1758.

Zhai, X., Mustafa, B., Kolesnikov, A., and Beyer, L. (2023). Sigmoid loss for language image pre-training.

Zhu, C., Tan, Y., Yang, S., Miao, J., Zhu, J., Huang, H., Yao, D., and Luo, C. (2024). Temporal dynamic synchronous functional brain network for schizophrenia classification and lateralization analysis. IEEE Transactions on Medical Imaging, 43(12):4307–4318.