Automatic classification of Candida species using Raman spectroscopy and machine learning.

Area of Science:

• Biomedical Spectroscopy
• Computational Biology
• Medical Mycology

Background:

• Hospital-acquired infections by pathogenic microorganisms pose significant risks.
• Rapid identification and treatment are crucial to prevent fatalities and combat antibiotic resistance.
• Traditional diagnostic techniques can be slow and laborious.

Purpose of the Study:

• To explore the automatic identification of eleven species within the genus Candida using Raman spectroscopy and machine learning.
• To assess the potential of Raman spectroscopy for rapid and reliable fungal identification.
• To develop and optimize machine learning models for classifying Candida species based on spectral data.

Main Methods:

• Acquisition of Raman spectra from over 220 measurements of dried Candida species cultures using a 532 nm laser confocal microscope.
• Development of a spectral preprocessing methodology.
• Training and hyperparameter optimization of various machine learning and deep learning algorithms, including a 1-D Convolutional Neural Network (1-D CNN).

Main Results:

• Analysis of spectral data revealed potential for discriminating between pathogenic yeast species.
• A 1-D CNN model achieved over 80% overall accuracy in identifying eleven Candida species.
• The developed model demonstrated good generalization capabilities for spectral datasets.

Conclusions:

• Raman spectroscopy combined with machine learning offers a powerful, rapid, and reliable method for identifying clinically relevant Candida species.
• The 1-D CNN approach shows significant promise for automated fungal identification in healthcare settings.
• This technique can aid in timely treatment, potentially reducing the impact of invasive fungal infections and antimicrobial resistance.