Porto, June 26 (Lusa) – A computer vision tool that makes breast cancer diagnosis quicker and better quality, providing specialists with a second opinion, is under development in the northern Portuguese city of Porto.
Confirmation of a breast cancer diagnosis, through visual analysis of histological images, “is not a trivial task,” and “often raises doubts among specialists,” researcher Teresa Araújo, told Lusa.
“Physical and psychological fatigue of specialists,” along with “the limits of visual systems,” an a “potential lack of experience,” can lead to poor diagnoses and computer systems can make up for some of these problems, said the researcher from the Research Centre for Biomedical Engineering (C-BER) of the Institute of Systems and Computer Engineering (INESC TEC) in Porto.
According to Araújo, this tool makes it possible to analyse a tissue biopsy through a series of pigmented images, classifying them as normal tissue, benign lesion, localised carcinoma or invasive carcinoma.
To do this, the tools uses convoluted neural networks (systems with millions of parameters inspired by the human visual system), which have allowed advances in medical image analysis performance and reduced the need for knowledge of the diagnostic process by specialists, providing a more independent opinion.
According to Araújo the tool is not intended to replace human diagnosis, but rather to “promote critical thinking,” through a “completely objective second opinion,” based on data.
The study was part of the “NanoSTIMA – Macro-to-Nano Human Sensing: Towards Integrated Multimodal Health Monitoring and Analytics,” project funded by the Norte 2020 programme, via the European Regional Development Fund, and was recently published in science magazine ‘Plos One’.
The two-year project also involved researchers Guilherme Aresta, Eduardo Castro, José Rouco e Aurélio Campilho, from INESC TEC, as well as António Polónia, Cataria Eloy andPaulo Aguiar, of the Institute for Research and Innovation in Health of the University of Porto (i3S).
TYP/CA // CA