Use of artificial intelligence for the detection of infections caused by Candida spp.: a systematic review

Abstract

Introduction: Candida spp. is composed of more than 160 species present in nature. However, only about 20 of them pose a threat to humans today. Situations of imbalance can lead to the spread and, consequently, fungal infection in various regions of the body. In this sense, Artificial Intelligence (AI) has proven to be an important tool in the identification of these infectious conditions. This study aims to evaluate the main studies related to the use of AI in the identification of fungi of the genus Candida and/or in the diagnosis of candidemia. Methods: Systematic review was conducted in the Pubmed, Lilacs, Cochrane and Embase databases, according to the guidelines of the International Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA). Results: Of the 403 studies identified, 4 were selected for inclusion in the review. Each study employed different means to obtain data, including an image bank of clinical specimens captured by brightfield microscopy, Raman spectroscopy, E-Nose, and clinical and laboratory databases. Subsequently, these data were analyzed by AI's, highlighting InceptionV3, 1-D CNN, Inception Time and Random Forest, which obtained the best performances. Conclusion: From the analysis of the studies, it was noted that the machine learning tools ensured greater agility and cost reduction, compared to traditional methods, in the detection of Candida spp. infection. In addition, these techniques demonstrated higher than average accuracy, sensitivity, and specificity. Thus, it is evident that AI's are technologies of great potential for the identification of fungal infections by Candida species.