OOCYHISTDB: UM CONJUNTO DE DADOS DE IMAGENS HISTOLOGICAS PARA DETECCAO DE OVOCITOS

Abstract

Detecting oocytes is a relatively new area of researchin the computational field. In this context, wepresent OocyHistDB, a dataset of histological, highresolutionimages depicting the oocyte stages of theappropriately labeled and publicly available speciesCentropomus undecimalis. This dataset is intended tobecome a powerful resource for researchers by providing acommon reference for comparison, testing and evaluationof existing and future learning techniques. In this paper,we describe how the OocyHistDB was collected, organized,and tested as a deep learning technique for detecting oocytephases. The technique obtained promising results, achievingan accuracy rate of 83.0% for class VI - early vitellogenesis,and 97.7%, 95.3% and 66.8%, respectively of revocation,mAP@0.5 and mAP@0.95, for class VF - late vitellogenesis.