Multi-target fluorescence staining of bacteria smears enables rapid machine learning-assisted species classification

Area of Science:

• Microbiology
• Medical Diagnostics
• Machine Learning

Background:

• Rapid bacterial identification is critical for effective treatment of infections, particularly bloodstream infections, to reduce mortality.
• Current methods like MALDI-TOF MS and multiplex PCR are effective but require specialized equipment and expertise, limiting their use in resource-limited areas.

Purpose of the Study:

• To develop and validate a novel, accessible staining and imaging procedure for bacterial species identification.
• To train a machine learning model for accurate species classification using extracted image features.

Main Methods:

• A staining and imaging procedure using fluorescent dyes targeting bacterial intracellular structures and cell wall components was developed.
• Image analysis extracted relevant features for training a machine learning model.
• The method was tested on clinical isolates from 126 patients, evaluating classification performance using ROC curves.

Main Results:

• The machine learning model achieved high classification performance for common bacteria, with area under the ROC curve ranging from 0.8 (Klebsiella pneumoniae) to 1 (Pseudomonas aeruginosa).
• The method reliably identified unknown species not included in the training dataset.
• The approach demonstrated potential for identifying Enterobacterales and other species.

Conclusions:

• The developed staining and imaging method, combined with machine learning, provides a promising, accessible tool for rapid bacterial species identification.
• This technique has significant potential for clinical application, particularly in improving diagnostics in resource-limited settings.