Animal fertility is a widely studied topic to promote the breeding of animals that inherit characteristics from their progenitors. Especially in equines, there are highly valued mares used for different equestrian sports whose genetic material is very valuable. Equines are not selected for their reproductive characteristics like other production animals but for their phenotype or sport aptitudes. Uterine health is fundamental for a healthy gestation, so it is of utmost importance to know the state of the mare’s uterus. For this purpose, endometrial biopsies are performed, among other techniques. In these biopsies, pathologists can study the presence and disposition of the different structures in the endometrium and estimate the potential degree of fertility of the animal. This work integrates image processing and machine learning techniques to analyze endometrial biopsies from mares, with the aim of reducing sample evaluation times and providing quantitative data for diagnosis. Two models were used to segment the images: one for glands, adapted from a model pre-trained on human tissue, and one for fibrosis, trained on a database collected and labeled during the research. A learning-based color normalization technique was also applied to ensure the latter model’s robustness to sample variations.