Determination of canine reproductive cycle stage from diestrus, proestrus and anestrus vaginoscopic images through computer vision techniques

Classical clinical evaluations for determination of reproductive cycle stage of a bitch are based on vaginal cytology, evaluation of progesterone and LH hormone levels, vaginal and vulvar inspection, behavioral changes and history. These evaluations requires expertise, equipment and are relatively time consuming. Our purpose in this study is to device a faster and updated method to support other techniques. Therefore, the study is planned as, analysis of vaginoscopic images with related computer vision techniques. In this study, images collected from 9 different healthy female dogs at various cycle stages. Correct cycle stages detected with vaginal cytology. Images are collected with 7 mm diameter LED light source USB camera. Insertion of the camera is maintained with 3d printed tubes that varies 9 mm to 15 mm. Tubes are inserted in the vagina in order to guide the camera. Images are then stored for software analysis. The purpose of our image analysis method is to process an image and estimate the reproductive cycle stage with a confidence value. This is modelled as an instance of data classification problem, the method has to be standardized to data consisting correct stage information. This study is consist of 47 images from 3 different stages. The structured data group went through a series of software based processes and high accuracy aimed. One of the method paths used fort his study is segmenting the image based on color information, extracting histograms with local binary pattern (LBP) features and training with support vector machines (SVM) algorithm. The performance of these processes are calculated with cross validation technique and confusion matrix. By calculating Kappa inter rater reliability, agreement between two methods shown statistically and for all the moethods performed P<0.001 sensitivity obtained. Results shows that, with combinations of software based approaches it is possible to adopt vaginal image analysis technique to routine practical basis. With collecting more images it may help to build a detailed vaginoscopic image data set for veterinary gynecology field and also it may become possible to make reliable estimations. It may also help to the clinical practitioner, while making the differantial diagnose list, to detect the reproductive cycle stage of a female dog without requiring laboratory technical expertise and also detect some pathologies.