VISÃO COMPUTACIONAL APLICADA À DETECÇÃO DE ANAPLASMA PLATYS EM PLAQUETAS

Abstract

The manual identification of Anaplasma platys in platelets, essential for diagnosing canine anaplasmosis, is a time-consuming and error-prone process. This work proposes a machine learning model (YOLOv8) to automate the detection and classification of healthy and infected platelets. For this, a hybrid dataset (real images from the HV-IF Goiano - Campus Urutaí and the internet) was built, trained with an 80% train and 20% validation split. The results demonstrate the prototype's viability: graphical analysis indicated the model learned to identify patterns but revealed a tendency for false positives (classifying healthy platelets as diseased). It is concluded that the tool is promising for auxiliary diagnosis, but its precision is limited by the restricted dataset. For future work, dataset expansion is recommended to improve performance.