Abstract
Candidemia is a severe fungal bloodstream infection with high mortality rates, especially in immunocompromised patients. Traditional diagnostic methods, such as blood cultures, suffer from low sensitivity and long turnaround times, delaying targeted antifungal therapy. In this study, we explore the potential of electronic noses (e-noses) combined with artificial intelligence (AI) models to provide a rapid and accurate diagnosis of Candida infections directly from blood culture broth samples. Samples were analyzed using an e-nose, generating datasets that underwent preprocessing steps including outlier removal, cycle equalization, feature transformation, and optional oversampling to address species imbalances. To ensure clinical relevance, we employed a dual-scenario validation strategy, assessing both intrasample sensor stability and intersample biological generalization. We tested both traditional machine learning models and time-series classifiers, selecting models based on prior research and performance in similar tasks. The best-performing models were evaluated based on accuracy, precision, recall, F1-score, specificity, and computational efficiency. Results demonstrated that while the framework achieved >98% consistency in sensor stability tests, the support vector classifier (SVC) emerged as the most robust model for generalization, achieving statistical parity with complex time-series models like InceptionTime, but with significantly higher computational efficiency. This study highlights the feasibility of AI-enhanced e-noses for rapid Candida detection, offering a promising alternative to conventional diagnostics in clinical settings.
Funding Information
  • National Council for Scientific and Technological Development
  • Foundation for the Support of Science and Technology of the State of Pernambuco
  • Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES), Brazil (001)

This publication has 28 references indexed in Scilit: