Identifying non-O157 Shiga toxin-producing Escherichia coli (STEC) using deep learning methods with hyperspectral...

Area of Science:

• Microbiology
• Spectroscopy
• Data Science

Background:

• Non-O157 Shiga toxin-producing Escherichia coli (STEC) serogroups, termed the "Big-Six", are significant causes of human illness.
• Conventional identification methods for these STEC serogroups are complex and time-consuming.

Purpose of the Study:

• To develop and evaluate a label-free method for classifying
• Big-Six
• STEC serogroups at the cellular level using hyperspectral microscope imaging (HMI).
• To compare the performance of different spectral analysis and classification models for STEC identification.

Main Methods:

• Label-free hyperspectral microscope imaging (HMI) was used to capture spectral "fingerprints" of bacterial cells.
• Principal Component Analysis (PCA) and Stacked Auto-Encoder (SAE) were employed for spectral feature extraction.
• Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and Soft-Max Regression (SR) classifiers were evaluated using varying dataset sizes.

Main Results:

• SAE-based classification models significantly outperformed PCA-based models.
• SAE-LDA, SAE-SVM, and SAE-SR achieved accuracies of 93.5%, 94.9%, and 94.6%, respectively.
• The SAE-SR model, with an expanded dataset, reached an average accuracy of 94.9% for classifying
• Big-Six
• STEC serogroups, with individual serogroup accuracies ranging from 92.4% to 97.4%.

Conclusions:

• Hyperspectral microscope imaging (HMI) coupled with the SAE-SR classification model demonstrates high potential for accurate and efficient identification of
• Big-Six
• STEC serogroups.
• This label-free approach offers a viable alternative to conventional methods for bacterial serogroup classification.
• The study highlights the effectiveness of deep learning (SAE) for spectral feature extraction in microbial identification.