Automated detection of reproductive stages of female canine from vaginoscopic images

Vaginoscopy is commonly employed to assess the different phases of the reproductive cycle in female canines. The technique can be used to confirm the onset of estrus, which is the period during which the female is receptive to mating. During proestrus, the vaginal mucosa exhibits a characteristic appearance that can be observed through a vaginoscope. The vaginal mucosa is pink-colored and shows edema with longitudinal primary folds. During the later phase of proestrus, it becomes pale and exhibits secondary folds. During the oestrum, the mucosa becomes paler, edema is relieved, and characteristic folds known as ‘crenulations’ are exhibited. Evaluating reproductive stages through manual examination is a labor-intensive procedure demanding substantial training to minimize inaccuracies. Furthermore, the consistency of results obtained through this method by veterinarians can be limited. In this work, an automated classification of the various stages in the reproductive cycle of female canines using machine learning is proposed. 179 vaginoscopic images are used after enhancement using color transfer algorithm in l \(\alpha \beta\) $\alpha \beta$ space. Region of interest are segmented using fuzzy c-means clustering. Textural features extracted by the local binary pattern (LBP) method are exploited to classify regions of interest into different stages. The classification is performed using various methods, including Naive Bayes (NB), support vector machine (SVM), decision tree (DT), k-nearest neighbor (kNN), k-star, decision stump (DS), and rule learner (RL). The highest accuracy achieved is 85.26 \(\%\) $\%$ for kNN. Parameters namely sensitivity, F1 score, recall, and, specificity, are also obtained and found to be .853, .88, .853, and .79, respectively. This automated classification system will help veterinarians in identifying the reproductive stages of female canines thereby predicting the optimum mating period.