Topological Analysis Techniques for Improving Neural Network Performance in COVID-19 Detection Using Persistent Homology
Journal
Lecture Notes in Networks and Systems
Proceedings of the International Conference on Computer Science, Electronics and Industrial Engineering (CSEI 2023)
ISSN
2367-3370
2367-3389
Date Issued
2024
Author(s)
Israel Reyes
Karen Cáceres-Benítez
Ana Marcillo
Andre Vera
Carolina Cadena-Morejón
Fernando Villalba-Meneses
Paulina Vizcaíno-Imacaña
Diego Almeida-Galárraga
Andrés Tirado-Espín
Type
book-chapter
Abstract
In this study, we employ topological data analysis techniques on neural networks applied in COVID-19 detection, aiming to improve their predictive power. Leveraging the power of persistent homology, a mathematical tool for extracting topological features from intricate datasets, we turned chest X-ray images into a representation of the topological features. This representation was used to train and test the ability of neural networks to learn topological properties from images. We examine neural networks trained on chest radiographs containing both COVID-19 positive and negative cases. Our results suggest that, by identifying specific topological features correlated with COVID-19 detection, we may enhance the performance of the neural network models and analyze the underlying factors contributing to high accuracy rate of detection. The findings from this study contribute as exploratory advance in the field of medical imaging analysis and disease detection, showcasing the potential of topological analysis within neural networks.