Construction and selection of signs for non-invasive endometriosis diagnostics using machine learning

Endometriosis is a common but poorly understood disease. From the appearance of the first symptoms to the diagnosis, it sometimes takes more than ten years. There is still no treatment that can help to recover from it completely. Computational models can help in understanding the mechanisms by which immune, hormonal and vascular disorders manifest in endometriosis and complicate treatment. The study deals with the construction and selection of signs of endometriosis risk and the formation of a mathematical model using several machine learning algorithms. In this case, an analysis of the importance of the signs is carried out, in which a subset of the signs that do not degrade the performance characteristics of the model (accuracy, speed, stability of operation) is reduced. The method which enables the selection of signs for constructing a prognostic model based on a selector containing filtering methods of sign significance for a processed data set is proposed. Voting for the inclusion of the sign is carried out by means of the majority function. The quality of sign construction and selection in the subject area of non-invasive diagnosis of endometriosis was assessed by a mathematical risk prediction model for endometriosis based on logistic regression with 30 traits. Model performance was evaluated using common machine learning metrics: accuracy, sensitivity, specificity, F1-score, and area under the ROC curve. The best result was achieved with an AUC of 0.950. The material is valuable to medical professionals in cybernetics.

1. Benagiano G., Brosens I., Lippi D. The history of endometriosis. Gynecol Obstet Invest. 2014;78:1–9.

2. Wheeler J.M. Epidemiology of endometriosis-associated infertility. J Reprod Med. 1989;34:41–6.

3. Eskenazi B., Warner M.L. Epidemiology of endometriosis. Obstet Gynecol Clin North Am. 1997;24:235–58.

4. Nnoaham K.E., Hummelshoj L., Webster P., et al. Impact of endometriosis on quality of life and work productivity: a multicenter study across ten countries. Fertility Sterility. 2011;96(2):366–73.e8.

5. Dunselman G.A.J., Vermeulen N., Becker C., et al. ESHRE guideline: management of women with endometriosis. Human Reproduction. 2014;29:400–412.

6. Andres M.P., Borrelli G.M., Abrão M.S. Endometriosis classification according to pain symptoms: can the ASRM classification be improved? Best Practice & Research Clinical Obstetrics & Gynaecology. 2018;51:111–118.

7. Koga K., Takamura M., Fujii T., Osuga Y. Prevention of the recurrence of symptom and lesions after conservative surgery for endometriosis. Fertility Sterility. 2015;104:793–801.

8. Johnson N.P., Hummelshoj L., Adamson G.D., et al. World endometriosis society consensus on the classification of endometriosis. Human Reproduction. 2017;32:315–24.

9. Fauconnier, A. et al. Early identification of women with endometriosis by means of a simple patient-completed questionnaire screening tool: A diagnostic study. Fertility Sterility. 2021;116:1580–1589.

10. Eskenazi B. et al. Validation study of nonsurgical diagnosis of endometriosis. Fertility Sterility. 2001;76:929–935.

11. Chapron, C. et al. A new validated screening method for endometriosis diagnosis based on patient questionnaires. eClinicalMedicine. 2022;44:101263.

12. Abramovich S.G. Physiotherapy of pain: text. manual. Irkutsk: RIO IGMAPO; 2020. 72 p. (In Russ.).

13. Limanovskaya O.V., Alferieva T.I. Fundamentals of machine learning: a textbook. Ministry of Science and Higher Education of the Russian Federation. Yekaterinburg: Publishing House of the Urals. un-ta; 2020. 88 p. (In Russ.).

14. Linde V.A., Tatarova N.A. Endometrioses. Pathogenesis, clinical picture, diagnosis and treatment. M.: GEOTAR-Media; 2010. 192 p. (In Russ.).