Analysis of neural networks trained with evolutionary algorithms for the classification of breast cancer histological images

Author(s)

Miguel, Joao Pedro Miranda ; Neves, Leandro Alves ; Martins, Alessandro Santana ; do Nascimento, Marcelo Zanchetta ; Tosta, Thaina A. Azevedo

• Document type: Journal article
• Source: EXPERT SYSTEMS WITH APPLICATIONS; v. 231, p. 14-pg., 2023-06-16.

Abstract

Biopsy tests used in the identification and confirmation of breast cancer are time-consuming and complex. Thus, neural networks can be applied to aid specialists with a prone to be in local minima, which can be avoided using evolutionary algorithms. In that way, this work analyzed the methods of genetic algorithm, differential evolution, differential evolution with islands and migration, particle swarm optimization, and adaptive particle swarm optimization in the training of neural networks for the classification of histological images stained by hematoxylin-eosin. The differential evolution with islands and migration and particle swarm optimization presented the most promising results, with the first one reaching a maximum AUC of 0.70 for the color features, and the last one with a maximum AUC of 0.71 for the texture attributes, on the evaluated dataset. Through this proposal, we have an important contribution to breast cancer histological image classification that also allows the development of new studies in the future. (AU)

• FAPESP's process: 22/03020-1 - Normalization of H&E stain by autoencoders with analysis of ensemble learning for histological images
• Grantee: Thaína Aparecida Azevedo Tosta
• Support Opportunities: Research Grants - Initial Project