From ATR-FTIR Spectra to Visibility Graphs: An End-to-End GNN Pipeline for ASD Detection

• Lucas G. T. Araújo UFU
• Robinson Sabino-Silva UFU
• Murillo G. Carneiro UFU

DOI: https://doi.org/10.5753/sbcas.2026.21273

Resumo

Autism Spectrum Disorder (ASD) lacks objective biomarkers, with diagnosis relying on behavioural observation and taking years. Salivary ATR-FTIR spectroscopy offers a non-invasive molecular fingerprint, but prior graph-based methods depend on hand-crafted topological features. We propose an end-to-end GNN pipeline that encodes each spectrum as a windowed visibility graph with a five-dimensional node feature vector and evaluates five architectures under stratified group cross-validation. GCN achieves F1MH = 0.810 in cross-validation and GIN 0.71 on the held-out test, competitive with prior graph-based approaches without hand-crafted features, establishing end-to-end GNN classification of ATR-FTIR spectra as viable for ASD detection.

Referências

Araújo, L. G., Sabino-Silva, R., and Carneiro, M. G. (2024). High-level classification using complex networks for autism spectrum disorder detection. In Simpósio Brasileiro de Computação Aplicada à Saúde (SBCAS), pages 331–341. SBC.

Association, A. P. et al. (2014). DSM-5: Manual diagnóstico e estatístico de transtornos mentais. Artmed Editora.

BRASIL, M. d. S. (2021). Definição - transtorno do espectro autista (tea) na criança. Acessado em: 18/01/2025.

Caixeta, D. C., Carneiro, M. G., Rodrigues, R., Alves, D. C. T., Goulart, L. R., Cunha, T. M., Espindola, F. S., Vitorino, R., and Sabino-Silva, R. (2023). Salivary atr-ftir spectroscopy coupled with support vector machine classification for screening of type 2 diabetes mellitus. Diagnostics, 13(8):1396.

Filho, R. B. L., Sabino-Silva, R., and Carneiro, M. G. (2025). High-level classification based on meta-graph of visibility graphs for autism detection. In 2025 International Joint Conference on Neural Networks (IJCNN), pages 1–8.

Guimarães, M. (1985). Exames de laboratório: sensibilidade, especificidade, valor preditivo positivo. Revista da Sociedade Brasileira de Medicina Tropical, 18:117–120.

Hamilton, W. L., Ying, R., and Leskovec, J. (2018). Inductive representation learning on large graphs.

Kipf, T. N. and Welling, M. (2016). Semi-supervised classification with graph convolutional networks. arXiv preprint arXiv:1609.02907.

Lacasa, L., Luque, B., Ballesteros, F., Luque, J., and Nuno, J. C. (2008). From time series to complex networks: The visibility graph. Proceedings of the National Academy of Sciences, 105(13):4972–4975.

Lima-Filho, R. B. and Carneiro, M. G. (2023). Diagnóstico do câncer oral através da classificação de alto nível. In Anais Estendidos do XXIII Simpósio Brasileiro de Computação Aplicada à Saúde, pages 54–59. SBC.

Lima Filho, R. B., Fernandes, J. M., Ji, D., Zhao, L., Sabino-Silva, R., and Carneiro, M. G. (2024). High-level network-based detection of oral cancer from atr-ftir spectroscopy. In 2024 International Joint Conference on Neural Networks (IJCNN), pages 1–8. IEEE.

Oliveira, S. W., Cardoso-Sousa, L., Georjutti, R. P., Shimizu, J. F., Silva, S., Caixeta, D. C., Guevara-Vega, M., Cunha, T. M., Carneiro, M. G., Goulart, L. R., et al. (2023). Salivary detection of zika virus infection using atr-ftir spectroscopy coupled with machine learning algorithms and univariate analysis: A proof-of-concept animal study. Diagnostics, 13(8):1443.

Santos-Jr, A. P., Filho, A. C. M., Sabino-Silva, R., and Carneiro, M. G. (2023). Convolutional neural networks for the molecular detection of covid-19. In Brazilian Conference on Intelligent Systems, pages 51–62. Springer.

Silva, S. F. d. P. et al. (2020). Avaliação de biomarcadores salivares para diagnóstico de transtorno de espectro autista por espectroscopia atr-ftir.

Veličković, P., Cucurull, G., Casanova, A., Romero, A., Lio, P., and Bengio, Y. (2017). Graph attention networks. arXiv preprint arXiv:1710.10903.

Wu, L., Cui, P., Pei, J., Zhao, L., and Song, L. (2022). Graph neural networks. Springer.

Xu, K., Hu, W., Leskovec, J., and Jegelka, S. (2019). How powerful are graph neural networks?