Automation of Multi-Class Microscopy Image Classification Based on the Microorganisms Taxonomic Features Extraction.

Area of Science:

• Microbiology
• Computational Biology
• Machine Learning

Background:

• Accurate and efficient classification of microorganisms is crucial for diagnostics.
• Current methods, especially deep learning, can be computationally intensive and lack interpretability.
• A need exists for lightweight, interpretable models for microbial classification.

Purpose of the Study:

• To develop a unified, low-parameter automated machine learning approach for multi-class microorganism classification.
• To enable interpretable taxonomic descriptors by analyzing external geometric characteristics.
• To provide a computationally efficient and lightweight alternative to deep learning models.

Main Methods:

• Utilized automated machine learning focusing on external geometric characteristics (cell shape, colony organization, dynamic behavior).
• Developed a low-parameter model for fast inference on standard CPU hardware.
• Created and published an annotated dataset of four bacterial types for validation.

Main Results:

• Achieved high performance metrics: Precision = 0.910, Recall = 0.901, F1-score = 0.905.
• Demonstrated effectiveness for biomedical diagnostic tasks.
• Showcased performance comparable to state-of-the-art methods with superior efficiency.

Conclusions:

• The proposed lightweight, low-parameter method effectively classifies microorganisms with interpretable descriptors.
• The approach is suitable for resource-limited settings and offers significant computational advantages.
• This method advances automated microbial diagnostics through efficiency and interpretability.