Machine learning to identify clinically relevant Candida yeast species.

Area of Science:

• Medical Mycology
• Computational Biology
• Health Informatics

Background:

• Rising incidence of fungal infections, particularly Candida species.
• Challenges in rapid and accurate identification of multi-drug resistant Candida auris.
• Growing application of machine learning in healthcare and medical imaging.

Purpose of the Study:

• To evaluate the efficacy of six convolutional neural networks (CNNs) for identifying four clinically significant Candida species.
• To compare the performance of different CNN architectures using microscopy images.
• To determine the optimal machine learning approach for Candida species identification.

Main Methods:

• Acquisition of wet-mounted microscopy images of Candida species.
• Separation of images into single-cell, budding-cell, and cell-group categories.
• Application of six machine learning algorithms (custom CNN, VGG16, ResNet50, InceptionV3, EfficientNetB0, EfficientNetB7) for species prediction.

Main Results:

• InceptionV3 demonstrated superior performance in predicting Candida species from microscopy images.
• All models performed poorly on raw, unprocessed images but improved with single and budding cell images.
• InceptionV3 achieved high accuracy rates for identifying budding and single cells of C. albicans, C. auris, C. glabrata, and C. haemulonii.

Conclusions:

• Microscopy images from wet-mounted slides can be effectively utilized for rapid and accurate Candida yeast species identification using machine learning.
• The InceptionV3 model shows significant potential for clinical application in fungal diagnostics.
• Further development of machine learning models can enhance the speed and accuracy of identifying challenging fungal pathogens.