Machine learning for the diagnosis of early-stage diabetes using temporal glucose profiles


Machine learning shows remarkable success for recognizing patterns in data. Here, we apply machine learning (ML) for the diagnosis of early-stage diabetes, which is known as a challenging task in medicine. Blood glucose levels are tightly regulated by two counter-regulatory hormones, insulin and glucagon, and the failure of glucose homeostasis leads to a common metabolic disease, diabetes mellitus. It is a chronic disease that has a long latent period that complicates detection of the disease at an early stage. The vast majority of diabetes cases result from that diminished effectiveness of insulin action, and that insulin resistance modifies the temporal profile of blood glucose. Thus, we propose to use ML to detect subtle changes in the temporal pattern of the glucose concentration. Time series data on blood glucose with sufficient resolution is currently unavailable, so we confirm the proposal by using synthetic glucose profiles produced using a biophysical model that considers glucose regulation and hormone action. Multi-layered perceptrons, convolutional neural networks, and recurrent neural networks all identified the degree of insulin resistance with high accuracy above \(85\%\).

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2


  1. 1.

    P.V. Röder, B. Wu, Y. Liu, W. Han, Exp. Mol. Med. 48(3), e219 (2016)

    Article  Google Scholar 

  2. 2.

    L. von Bertalanffy, Science 111, 239 (1950)

    Article  Google Scholar 

  3. 3.

    D.A. Lang, D.R. Matthews, J. Peto, R.C. Turner, New Engl. J. Med. 301, 10237 (1979)

    Article  Google Scholar 

  4. 4.

    American Diabetes Association, Diabetes Care 41(Suppl. 1), S13 (2018)

    Article  Google Scholar 

  5. 5.

    American Diabetes Association, Diabetes Care 42(Suppl. 1), S13–S28 (2019)

    Article  Google Scholar 

  6. 6.

    J.C. Seidell, Br. J. Nutr. 83(Suppl. 1), S5 (2000)

    Article  Google Scholar 

  7. 7.

    J.C. Seidell, Horm. Metab. Res. 21, 155 (1998)

    Google Scholar 

  8. 8.

    G.M. Reaven, Physiol. Rev. 75(3), 473 (1995)

    Article  Google Scholar 

  9. 9.

    S. Larabi-Marie-Sainte, L. Aburahmah, R. Almohaini, T. Saba, Appl. Sci. 9(21), 4604 (2019)

    Article  Google Scholar 

  10. 10.

    I. Kavakiotis, O. Tsave, A. Salifoglou, N. Maglaveras, I. Vlahavas, I. Chouvarda, Comput. Struct. Biotechnol. J. 15, 104 (2017)

    Article  Google Scholar 

  11. 11.

    D. Park, T. Song, D. Hoang et al., Sci. Rep. 7, 1602 (2017)

    Article  ADS  Google Scholar 

  12. 12.

    T. Song, J. Jo, Phys. Biol. 16, 051001 (2019)

    Article  ADS  Google Scholar 

  13. 13.

    G.M. Grodsky, J. Clin. Investig. 51, 2047 (1972)

    Article  Google Scholar 

  14. 14.

    M. Komatsu, M. Takei, H. Ishii, Y. Sato, J. Diabetes Investig. 4, 511 (2013)

    Article  Google Scholar 

  15. 15.

    I.J. Stamper, X. Wang, J. Theor. Biol. 318, 210 (2013)

    Article  Google Scholar 

  16. 16.

    P. Palumbo, S. Ditlevsen, A. Bertuzzi, A.D. Gaetano, Math. Biolsci. 244, 69 (2013)

    Article  Google Scholar 

  17. 17.

    I.M. Tolic, E. Mosekilde, J. Sturisa, J. Theor. Biol. 207, 361 (2000)

    Article  Google Scholar 

  18. 18.

    J. Li, Y. Kuang, C.C. Mason, J. Theor. Biol. 242, 722 (2006)

    Article  Google Scholar 

  19. 19.

    B. Lee, T. Song, K. Lee, J. Kim, S. Han, P.-O. Berggren et al., PLoS One 12(2), e0172901 (2017)

    Article  Google Scholar 

  20. 20.

    T. Song, H. Kim, S.-W. Son, J. Jo, Phys. Rev. E 101, 022613 (2020)

    Article  ADS  Google Scholar 

  21. 21.

    B. Lee, T. Song, K. Lee, J. Kim, P.-O. Berggren, S.H. Ryu et al., PLoS One 12(8), e0183569 (2017)

    Article  Google Scholar 

  22. 22.

    We set the amplitude modulation functions \(f_{\alpha }(G)=[1-\tanh ((G-G_0)/\delta G)]/2\), \(f_{\beta }(G)=[1+\tanh ((G-G_0)/\delta G)]/2\), and \(f_{\delta }(G)=[1+\tanh ((G-G_0+\Delta G)/\delta G)]/4\) and the phase modulation functions \(\mu _{\alpha }(G)=\mu \cdot (G-G_0)\) and \(\mu _{\beta }(G)= \mu _{\delta }(G)=-\mu \cdot (G-G_0)\). Here, we used the parameters, \(G_0=6.3~\text{mM}\), \(\Delta G=4.5~\text{ mM }\), \(\delta G=1.8~\text{ mM }\), and \(\mu =0.1~\text{ mM}^{-1}\)

  23. 23.

    P. Esling, C. Agon, ACM Comput. Surv. 45(1), 1 (2012)

    Article  Google Scholar 

  24. 24.

    A. Bevilacqua, K. MacDonald, A. Rangarej, V. Widjaya, B. Caulfield, T. Kechadi, Human activity recognition with convolutional neural networks, in Machine Learning and Knowledge Discovery in Databases. ECML PKDD 2018. Lecture Notes in Computer Science, vol. 11053, ed. by U. Brefeld et al. (Springer, Cham, 2019)

  25. 25.

    A. Craik, Y. He, J.L. Contreras-Vidal, J. Neural Eng. 16(3), 031001 (2019)

    Article  ADS  Google Scholar 

  26. 26.

    G. Hinton et al., IEEE Signal Process. Mag. 29(6), 82 (2012)

    Article  ADS  Google Scholar 

  27. 27.

    A. Mohamed, G. Hinton, G. Penn, in 2012 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Kyoto (2012), pp. 4273–4276

  28. 28.

    A. Krizhevsky, I. Sutskever, G. Hinton, Advances in Neural Information Processing Systems (NIPS 2012), pp. 1097–1105

  29. 29.

    H.I. Fawaz, G. Forestier, J. Weber, L. Idoumghar, P.-A. Muller, Data. Min. Knowl. Discov. 33, 917 (2019)

    MathSciNet  Article  Google Scholar 

  30. 30.

    A. Sherstinsky (2018). arXiv:1808.03314

  31. 31.

    We used a tangent hyperbolic activation function for the LSTM given the input shape \((100\times 5)\), because its long segment dimension caused the weight to explode when we used ReLU

  32. 32.

    D.P. Kingma, B. Jimmy (2014). arXiv:1412.6980

  33. 33.

    F.A. Gers, J. Schmidhuber, F. Cummins, in Ninth International Conference on Artificial Neural Networks ICANN 99 (Conf. Publ. No. 470) (1999)

  34. 34.

    K. Cho et al. (2014). arXiv:1406.1078

  35. 35.

    Chollet, François et al., Keras, 2015. Software available from

  36. 36.

    M. Abadi et al., TensorFlow: large-scale machine learning on heterogeneous systems (2015). Software available from

  37. 37.

    Z. Mian, K.L. Hermayer, A. Jenkins, Am. J. Med. Sci. 358, 332 (2019)

    Article  Google Scholar 

  38. 38.

    M. Vettoretti, G. Cappon, G. Acciaroli, A. Facchinetti, G. Sparacino, J. Diabetes Sci. Technol. 12, 1064 (2018)

    Article  Google Scholar 

  39. 39.

    G. Cappon, M. Vettoretti, G. Sparacino, A. Facchinetti, Diabetes Metab. J. 43, 383 (2019)

    Article  Google Scholar 

Download references


This research was supported by the POSTECH (T.S.) and Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, NRF-2019R1F1A1052916, (J.J.) and NRF-2017R1D1A1B03034600 (T.S.).

Author information



Corresponding author

Correspondence to Taegeun Song.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Lee, W.S., Jo, J. & Song, T. Machine learning for the diagnosis of early-stage diabetes using temporal glucose profiles. J. Korean Phys. Soc. (2021).

Download citation


  • Machine learning
  • Diagnosing diabetes
  • Insulin resistance