SEGMENTAÇÃO DO CORPO LÚTEO EM IMAGENS DE ULTRASSOM DOPPLER POR MEIO DE CNN COM FINALIDADE DE VIABILIZAR O DIAGNÓSTICO PREOCE DE PRENHEZ EM VACAS

Abstract

The early determination of pregnancy in cows using ultrasound can generate great savings and increase herd production. Observing the evidence that perfusion in the corpus luteum provides a good parameter to determine early pregnancy, this study aims to perform the segmentation of images from Doppler applied to the reproductive system of bovines, in order to enable early diagnosis of pregnancy, since the calculation of the perfusion area in Doppler ultrasound is trivial in face of a segmented corpus luteum. To perform the segmentation task, a CNN with UNet architecture was developed, resulting in an accuracy of 0.8847 in the set of test images.