A Complete Transfer Learning-Based Pipeline for Discriminating Between Select Pathogenic Yeasts from Microscopy...

Area of Science:

• Medical Mycology
• Computational Biology
• Clinical Diagnostics

Background:

• Pathogenic yeasts, including drug-resistant *Candida auris*, pose significant healthcare risks, particularly to immunocompromised individuals.
• Accurate and rapid yeast identification is crucial for effective treatment, especially in resource-limited settings.
• Current diagnostic methods can be slow, expensive, and complex, necessitating innovative solutions.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN)-based approach for the accurate classification of pathogenic yeast species using microscopy images.
• To create a cost-effective and accessible diagnostic tool for yeast identification.
• To improve diagnostic response times in clinical environments.

Main Methods:

• A convolutional neural network (CNN) model was trained using transfer learning on microscopy images of six pathogenic yeast species.
• The model was evaluated for classification accuracy at both the patch and whole image levels.
• Performance metrics included accuracy and misclassification rates across different species.

Main Results:

• The CNN model achieved high classification accuracy, reaching 93.91% at the patch level and 99.09% at the whole image level.
• The best-performing model demonstrated low misclassification rates among the tested yeast species.
• The developed pipeline provides a streamlined and efficient method for yeast identification.

Conclusions:

• The CNN-based approach offers a highly accurate and efficient method for identifying pathogenic yeasts from microscopy images.
• This AI-driven tool presents a cost-effective and accessible alternative to traditional molecular diagnostic methods.
• The study highlights the potential of machine learning in enhancing rapid diagnostics for fungal infections in healthcare settings.