Espectroscopia de aprendizado de máquina aplicada ao diagnóstico dos efeitos da covid-19 na reprodução humana

Abstract

Viral infections, including COVID-19, cause adverse effects on the male reproductive system and may lead to fertility issues, even in patients who have already recovered from the disease. However, the methods for human semen analysis recommended by the World Health Organization (WHO) are usually not able to identify significant differences in the sperm parameters, such as morphology, motility, and sperm count, of patients recovered from COVID-19 compared to those who have never contracted the disease. Therefore, we propose a novel method capable of distinguishing seminal samples from individuals who have never contracted COVID-19 from those who have recovered from the disease. Thus, two approaches involving the application of machine learning were developed. The first one involves a more in-depth analysis of the sperm parameters in patients’ ejaculate, while the second adopts spectroscopic measurements, specifically ellipsometry, to further characterize the same semen samples. In both methodologies, machine learning was used to classify the samples employing their measured data. Different classification algorithms (KNN, LR, SVM, MLP, DT e XGBoost) were trained, optimized, and validated to perform this task. The best result achieved with the first approach, using the SVM algorithm, demonstrated good performance, with an accuracy of 77%. By conducting an interpretability evaluation, we concluded that progressive motility and sperm immotility are the main parameters for distinguishing the samples of each class. The second method, using ellipsometry measurements, showed excellent performance with an accuracy of 90%. In this sense, it is evident that the methodology presented here is effective, being capable of detecting differences in the analyzed samples and classifying them with precision. Therefore, this study demonstrates the applicability of a tool with the potential to be used by urologists and fertility clinics to evaluate sperm parameters and their implications for male infertility. Furthermore, it paves the way for additional research to deepen and advance this still underexplored field.