Imbalanced Datasets and Bias in Artificial Intelligence: Influence of Sex and Age for COVID-19 Screening

EasyChair Preprint 13711

Abstract

In this study, we examined eleven imbalance scenarios in which COVID-19 patients were present in varying proportions for the sex analysis and six scenarios in which the age factor was trained using only one particular age range. Three distinct methods for automatically detecting COVID-19 were employed in each study: (I) COVID-19 VS Normal, (II) COVID-19 vs Pneumonia, and (III) Non-COVID-19 VS COVID-19. Two representative public chest X-ray datasets were used to validate the current work, enabling a trustworthy analysis to aid in clinical decision-making. The findings of the sex-related analysis show that this element has a minor impact on the COVID-19 deep learning-based systems, but not enough to significantly degrade the system. Age was shown to be influencing the system more consistently in the age-related study because it was present in every scenario that was taken into consideration.

Keyphrases: COVID-19 screening, Chest X-rays, data analysis, deep learning

@booklet{EasyChair:13711,
  author    = {Lorena Álvarez and Joaquim de Moura and Jorge Novo and Marcos Ortega},
  title     = {Imbalanced Datasets and Bias in Artificial Intelligence: Influence of Sex and Age for COVID-19 Screening},
  howpublished = {EasyChair Preprint 13711},
  year      = {EasyChair, 2024}}