Prediction of gestational status nineteen days after artificial insemination using color ultrasonography and machine learning strategies in beef heifers.

This article aims to develop a predictive model of gestational status nineteen days after artificial insemination, using color ultrasonography and machine learning strategies in beef heifers. One hundred heifers were included in the study and on day 0 of the experiment underwent fixed-time artificial insemination. In addition, live weight, body condition score, uterine diameter, estradiol and progesterone concentration were determined. On day nineteen, the presence and area of the corpus luteum, area of the cavity of the corpus luteum, estradiol and progesterone concentrations, and vascularized area of the corpus luteum and blood flow of the corpus luteum were determined. Pregnancy diagnosis was made on days 35. To identify the model predictive capacity, we implemented a machine learning strategy, specifically a Random Forest Classifier. Two models were evaluated, the complete and reduced models. Model performances were quantified with the accuracy, specificity, sensitivity, precision, error and area under the receiver operating characteristic curve (AUC). Complete model produced accuracy of 75.3%, sensitivity of 82.5%, specificity of 66.6%, precision of 75%, error of 24.6% and AUC of 0.97. Reduced Model produced accuracy of 78%, sensitivity 87.5%, specificity 66.6%, precision 76%, error of 21.9% and AUC of 0.95. In conclusion, the developed models reasonably predicted gestational status. While these findings should be interpreted as a tool under development and require increased sample size and external validation to confirm these preliminary observations, they offer an initial perspective on the importance of using machine learning strategies that integrate reproductive biotechnology, precision livestock farming, and computational analysis applied to bovine fertility.