Rapid Salmonella Serovar Classification Using AI-Enabled Hyperspectral Microscopy with Enhanced Data Preprocessing...

Area of Science:

• Microbiology
• Spectroscopy
• Artificial Intelligence

Background:

• Traditional Salmonella serovar identification is time-consuming and resource-intensive, relying on multiple enrichment steps and selective media.
• Rapid and accurate identification of Salmonella serovars is crucial for food safety and public health.
• Existing methods often lack the speed and efficiency required for high-throughput screening.

Purpose of the Study:

• To develop and validate a rapid, culture-independent method for Salmonella serovar identification using AI and hyperspectral microscopy.
• To compare the performance of spectral, image, and multimodal AI-based classification approaches.
• To demonstrate the potential of AI-driven hyperspectral imaging for streamlining bacterial identification workflows.

Main Methods:

• Hyperspectral data were collected from five Salmonella serovars (Enteritidis, Infantis, Kentucky, Johannesburg, 4,[5],12:i:-) using a culture-independent approach.
• Data analysis involved parallel spectral (manual feature selection vs. PCA) and image (CNN) branches, followed by multimodal fusion.
• Machine learning models (k-NN, SVM, Random Forest, MLP) were employed for spectral classification, while CNNs were used for image classification.

Main Results:

• Principal Component Analysis (PCA)-derived spectral features combined with machine learning achieved 81.1% accuracy, outperforming manual feature selection.
• Convolutional Neural Network (CNN) classification using image features alone resulted in lower accuracy (57.3%) for serovar discrimination.
• Multimodal fusion of spectral and image features significantly improved classification accuracy to 82.4% on an unseen test set, while reducing overfitting.

Conclusions:

• AI-enabled hyperspectral microscopy offers a rapid and effective alternative to traditional Salmonella serovar identification methods.
• Multimodal fusion of spectral and image data provides superior classification performance compared to using either data type alone.
• This approach has the potential to significantly streamline Salmonella serovar identification workflows in diagnostic and research settings.