Diagnosis of the protozoan Cryptosporidium spp. via digital image analysis

Author(s)

Saulo Hudson Néry Loiola

• Document type: Master's Dissertation
• Press: Campinas, SP.
• Institution: Universidade Estadual de Campinas (UNICAMP). Faculdade de Ciências Médicas
• Defense date: 2021-02-12

Examining board members

Jancarlo Ferreira Gomes; Katia Denise Saraiva Bresciani; Aline do Nascimento Benitez

• Advisor: Jancarlo Ferreira Gomes

Abstract

The diagnosis of Cryptosporidium spp. in fecal samples by detecting oocysts on slides under a conventional light microscope requires more practical, modern, and effective approaches and protocols, since cryptosporidiosis is a zoonosis of interest in Public Health, especially in countries with poor socioeconomic conditions and health infrastructure. We propose a step forward for improving the diagnosis of these protozoa with an emphasis on application in the Automated Diagnosis of Gastrintestinal Parasites System (DAPI).To this end, we present a parasitological technique that results in slides with a satisfactory concentration of oocysts, reasonable elimination of impurities, and temporary staining using a solution of trichrome modified by Melvin and Brooke at a concentration of 25%, as well as a computer program specially developed for diagnosing these protozoa, with the following sequence of algorithms: segmentation in superpixels, clustering of patches, filtering based on the typical size of the oocysts, and finally, application of a convolutional neural network. These algorithms point to patches with the potential for containing the objects of interest (oocysts), with 81.3% accuracy and 94.1% recall in the final stage. We significantly reduced the factors jeopardizing the success of the qualitative identification of this etiological agent, which are: impossibility of precise morphometry, refringence of the oocysts in temporary preparations, absence of contrast between parasites and other organic structures or not, satisfactory concentration of oocysts, clear field of view, and reduction in the experts¿ fatigue. This study allows to take a step forward towards the automated diagnosis of Cryptospodidium spp., which may be applicable in the remote training of health professionals, helping meet the high demands with reduced costs compared to serological and molecular methods, making its implementation in Public Health programs viable. (AU)

• FAPESP's process: 18/21204-7 - Diagnosis of Cryptosporidium spp. by automated image analysis
• Grantee: Saulo Hudson Nery Loiola
• Support Opportunities: Scholarships in Brazil - Master