Comparing machine learning and deep learning regression frameworks for accurate prediction of dielectrophoretic...

Area of Science:

• Biomedical Engineering
• Microfluidics
• Machine Learning

Background:

• Dielectrophoretic (DEP) force is crucial for manipulating microparticles in Lab-on-Chip devices.
• Accurate quantification of DEP force is essential for optimizing device performance.
• Existing methods for DEP force estimation can be complex and time-consuming.

Purpose of the Study:

• To develop and validate an intelligent sensing framework for precise quantification of dielectrophoretic force.
• To explore the efficacy of various Machine Learning (ML) and Deep Learning (DL) models in estimating DEP force.
• To establish a baseline for applying DL in DEP-assisted Lab-on-Chip sensing devices.

Main Methods:

• Utilized a textile electrode-based DEP sensing device to capture images of microparticle alignment.
• Developed deep regression models using modified ML and Convolutional Neural Network (CNN) architectures (AlexNet, ResNet-50, MobileNetV2, GoogLeNet).
• Correlated pearl chain alignment patterns of yeast cells and polystyrene microbeads to DEP force using various ML/DL models and evaluated performance using metrics like RMSE.

Main Results:

• ResNet-50 with RMSPROP achieved the best performance for yeast cells with a validation RMSE of 0.0918.
• AlexNet with ADAM optimizer demonstrated the best performance for microbeads with a validation RMSE of 0.1745.
• The framework showed strong prediction accuracy and generalization ability validated by experimental results.

Conclusions:

• The developed ML/DL framework accurately quantifies dielectrophoretic force in a textile electrode-based DEP sensing device.
• Deep learning models, particularly ResNet-50 and AlexNet, show significant promise for DEP force estimation.
• This work provides a foundational approach for future advancements in DL-driven Lab-on-Chip devices.