Detecção Automática de Glomérulos em Imagens Histológicas Renais Digitais
Jonathan Moreira Cardozo Rehem, Fúlvia Angelo, Michele, Washington L. C. dos‐Santos · Zenodo (CERN European Organization for Nuclear Research) · 2019
Glomerulopathies are kidney diseases that affects thousands people in Brazil and in the entire world, this number is growing constantly. Technological evolution has allowed the digitalization of histological slides, making possible the development of computational tools that can assist the work of pathologists. This work aims to investigate if traditional machine learning techniques are efficient in the automatic detection of glomeruli, when applied over a proper dataset, composed by images captured by a different and heterogeneous method. For this, were used SVM (Support Vector Machines) classifier, mrcLBP (Multi Radial Color Local Binary Pattern) feature extractor, Sliding Window scan and Pyramid Image. This work achieved recall of 0.595, precision of 0.981 and 0.741 of F1 Score. This paper demonstrates the obstacles encountered and contributes to the advances needed to achieve this goal.