Attention to FTIR: A Transformer Architecture for FTIR Spectra Classification in Oral Cancer Diagnosis

• Lucas S. Procópio UFU
• Robinson S. da Silva UFU
• Murilo G. Carneiro UFU
• Paulo D. Souza UFU

DOI: https://doi.org/10.5753/sbcas_estendido.2026.21597

Resumo

Fourier-transform infrared (FTIR) spectroscopy is a promising non-invasive technique for oral cancer diagnosis, whose high mortality is largely attributable to late-stage detection. This work proposes BioSpectralFormer (BSF), a Transformer-based architecture with two attention types for classification of salivary FTIR spectra. Evaluated on real spectra data under stratified 10-fold cross-validation against seven baselines, BSF achieved competitive balanced accuracy (Mean(SE,SP) = 0.67 ± 0.15) with high sensitivity (0.82 ± 0.20), operating in the same statistical tier as state-of-the-art methods. Moreover, attention map analysis corroborated established oral cancer biomarkers, including Amide I and lipid C-H stretching, suggesting biologically feature learning.

Referências

Balan, V. et al. (2019). Vibrational spectroscopy fingerprinting in medicine: from molecular to clinical practice. Materials, 12(18):2884.

Chen, T. and Guestrin, C. (2016). Xgboost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, page 785–794. ACM.

Gorishniy, Y., Kotelnikov, A., and Babenko, A. (2025). Tabm: Advancing tabular deep learning with parameter-efficient ensembling. In International Conference on Learning Representations, volume 2025, pages 77899–77935.

Hollmann, N. et al. (2022). Tabpfn: A transformer that solves small tabular classification problems in a second. arXiv preprint arXiv:2207.01848.

Holzmüller, D., Grinsztajn, L., and Steinwart, I. (2024). Better by default: Strong pre-tuned mlps and boosted trees on tabular data.

Huang, S., Jin, Y., Jin, W., and Mu, Y. (2025). Analytical-chemistry-informed transformer for infrared spectra modeling. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 17440–17448.

Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., Ye, Q., and Liu, T.-Y. (2017). Lightgbm: A highly efficient gradient boosting decision tree. In Advances in Neural Information Processing Systems, volume 30.

Khosla, P. et al. (2020). Supervised contrastive learning. In NeurIPS.

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.

Prokhorenkova, L., Gusev, G., Vorobev, A., Dorogush, A. V., and Gulin, A. (2018). Cat-boost: unbiased boosting with categorical features. In Advances in Neural Information Processing Systems, volume 31. Curran Associates, Inc.

Shree, P., Aggarwal, Y., Kumar, M., Majhee, L., Singh, N. N., Prakash, O., Chandra, A., Mahuli, S. A., Shamsi, S., and Rai, A. (2024). Saliva based diagnostic prediction of oral squamous cell carcinoma using FTIR spectroscopy. Indian J Otolaryngol Head Neck Surg, 76(3):2282–2289.

Su, K.-Y. and Lee, W.-L. (2020). Fourier transform infrared spectroscopy as a cancer screening and diagnostic tool: A review and prospects. Cancers, 12(1).

Swaminathan, D., George, N. A., Thomas, S., and Iype, E. M. (2024). Factors associated with delay in diagnosis of oral cancers. Cancer Treat. Res. Commun, 40:100831.

Yang, L. et al. (2022). Deep learning-based fourier transform infrared spectroscopy for identifying esophageal squamous cell carcinoma. Spectrochimica Acta Part A, 271:120891.

Zhang, Y. et al. (2024). Fcg-former: Functional group identification in ftir spectra using transformers. Analytical Chemistry, 96:7890–7899.