Repository logo

Classification of bovine reproductive cycle phases using ultrasound-detected features



Journal Title

Journal ISSN

Volume Title




Degree Level



With the combination of computer-assisted image analysis and ultrasonographic imaging technology, it has been possible to study and increase the knowledge in different areas of medicine. Studies of ovarian development in female mammals using ultrasonography have shown a relationship between the day in the estrous cycle and the main structures of the ovary. Ultrasound images of bovine ovaries were used to determine whether ultrasound-detected features can automatically determine the phase in the estrous cycle based on a single day's ultrasound examination of the ovaries. Five ultrasound-detected features of the bovine ovaries were used to determine the phase in the estrous cycle: (1) size of the dominant follicle; (2) size of the first subordinate follicle; (3) size of the second subordinate follicle; (4) size of the corpus luteum and (5) number of subordinate follicles with size ≥ 2mm. The collection of ultrasound images used for this study was formed by a group of 45 pairs of ovaries (left and right) which were imaged on day 3, day ⋍10 and day ≥ 17 of the estrous cycle corresponding to the metestrus, diestrus and proestrus phases respectively.Four different experiments were performed to test the hypothesis. For experiments 1, 2 and 3 the bovine ovaries were classified into three different classes: day 3 of wave 1 (D3W1), day 1 of wave 2 (D1W2) and day 17 or higher (D ≥17) that were related to the follicular development of the ovary and the estrous cycle phases as: metestrus, diestrus and proestrus respectively. For experiment 4 the bovine ovaries were classified into four classes: D3W1, D6W1, D1W2 and D ≥17. The additional class (D6W1: day 6 of wave 1) was incorporated to represent the early-diestrus phase in the estrous cycle.Two classifiers were implemented for all experiments and their performances compared: a decision tree classifier and a naive Bayes classifier. The decision tree classifier had the best performance with a classification rate of 100% for experiments 1, 2 and 3, giving a rather simple decision tree which used only two features to make a classification: size of the dominant follicle and size of the corpus luteum, suggesting these are key features in distinguishing between phases in the estrous cycle giving the most relevant information. The naive Bayes had a classification rate of 86.36% for experiment 1, 95.55% for experiment 2 and 90% for experiment 3. The results of this study supported the hypothesis that by using ultrasound detected features of bovine ovaries we can determine automatically the stage in the estrous cycle based on a single day's examination.



medical imaging, artificial intelligence, classification methods, bovine ovaries



Master of Science (M.Sc.)


Computer Science


Computer Science


Part Of