%0 Conference Proceedings
%4 sid.inpe.br/sibgrapi/2021/09.16.18.04
%2 sid.inpe.br/sibgrapi/2021/09.16.18.04.05
%T PathoSpotter-Search: A Content-Based Image Retrieval Tool for Nephropathology
%D 2021
%A Aguiar, Ellen,
%A Calumby, Rodrigo,
%A Oliveira, Luciano,
%A Santos, Washington,
%A Duarte, Angelo,
%@affiliation Universidade Estadual de Feira de Santana
%@affiliation Universidade Estadual de Feira de Santana
%@affiliation Universidade Federal da Bahia
%@affiliation Fundação Oswaldo Cruz
%@affiliation Universidade Estadual de Feira de Santana
%E Paiva, Afonso,
%E Menotti, David,
%E Baranoski, Gladimir V. G.,
%E Proença, Hugo Pedro,
%E Junior, Antonio Lopes Apolinario,
%E Papa, João Paulo,
%E Pagliosa, Paulo,
%E dos Santos, Thiago Oliveira,
%E e Sá, Asla Medeiros,
%E da Silveira, Thiago Lopes Trugillo,
%E Brazil, Emilio Vital,
%E Ponti, Moacir A.,
%E Fernandes, Leandro A. F.,
%E Avila, Sandra,
%B Conference on Graphics, Patterns and Images, 34 (SIBGRAPI)
%C Gramado, RS, Brazil (virtual)
%8 18-22 Oct. 2021
%I Sociedade Brasileira de Computação
%J Porto Alegre
%S Proceedings
%K Computational Pathology, Content Based Image Retrieval, Nephropathology, Convolutional Networks.
%X Nephropathologists typically organizes their repository of digital images of kidney biopsies in such a way that it is difficult to retrieve cases that have images similar to a picture under analysis. Having this in mind, we initiated the development of PathoSpotter-Search, a Content-Based Image Retrieval system for images of kidney biopsies. The system operates as a cloud service to avoid the need to install any software on the pathologists computer. Our approach combines a feature extractor followed by a similarity score calculator. We evaluated convolutional network (CN) architectures (VGG-16 (original and fine-tuned) and Inception-ResNet, and a network used in the proprietary classifier for glomerular hypercellularity), combined with Cosine and Euclidean distances as similarity scores. The first results have shown that the CN of the VGG16 combined with cosine distance yielded the best performance (precision ≈ 53%). To assess the usability and functionality of the PathoSpotter-Search as a cloud service, the system was tested by nephropathologists and proved to be useful as a tool for retrieving similar images from their local repositories. Currently, we are working to improve the system precision to at least 70%, and evaluating strategies to retrieve similar images based on segments or tiles of the query image.
%@language en
%3 PathoSpotter-Search_ A Content-Based Image Retrieval Tool for Nephropathology.pdf