Aproximación al desempe?o operativo de un modelo de redes neuronales en el diagnóstico de la infertilidad masculina

摘要

Objective: To determine the ability of supervised neuronal networks at making the appropriate classification of fertile and infertile men using conventional seminal parameters. Materials and methods: Cross-sectional study assembled on the database of the Universidad de Antioquia Reproduction Group, with a selection of men experiencing reproductive problems within the previous 12 months, and men with a history of having had children. Convenience sampling. The data considered were age, time of sexual abstinence, ejaculate volume, pH, percentage of sperm motility, viability and concentration. Using a supervised neuronal network, a training model and a validation model were created. Results: Overall, 204 men were included, 129 for the training model, 35 for validation, 40 for testing the model and 25 for external validation. The neuronal network model made the correct classification of 90% of the subjects with reproductive problems, and 91% of the fertile subjects. In the validation model, the neuronal network made the correct classification of 40% of the subjects with reproductive problems, and 100% of the fertile subjects. Conclusion: Neuronal networks emerge as a technology that may prove to be valuable for the study of male infertility. More rigorous evaluations are required in order to determine their true usefulness in the study of infertile couples.