Electro-Optical Classification of Pollen Grains via Microfluidics and Machine Learning

Area of Science:

• Biotechnology
• Sensor Technology
• Machine Learning

Background:

• Traditional pollen analysis methods are costly, labor-intensive, and prone to errors.
• Accurate, automated, and label-free pollen identification is crucial for aerobiological monitoring and agriculture.

Purpose of the Study:

• To develop a novel multimodal approach for high-throughput, automated pollen grain classification.
• To integrate electrical sensing and optical imaging with machine learning for enhanced accuracy.

Main Methods:

• A microfluidic chip was used to analyze pollen grains at a rate of 150 grains per second.
• Electrical signals and synchronized optical images were captured and processed by independent machine learning classifiers.
• A final classification outcome was achieved by combining the predictions from both classifiers.

Main Results:

• The multimodal approach achieved an average balanced accuracy of 84.2% and an overall accuracy of 88.3% across eight pollen classes.
• Individual electrical and optical classifiers showed accuracies of 78.7% and 76.7% (balanced) and 82.8% and 84.1% (overall), respectively.
• The combined multimodal classification significantly outperformed single-modality analyses.

Conclusions:

• The multimodal approach offers superior classification performance compared to methods relying solely on electrical or optical data.
• This methodology enables automated multimodal palynology and has potential applications in diagnostics and cell therapy for label-free cell identification.