Neural Network-Based Filter Design for Compressive Raman Classification of Cells.

Area of Science:

• Biotechnology
• Spectroscopy
• Artificial Intelligence

Background:

• Cell-based therapies require advanced characterization methods for production and quality control.
• Current label-free techniques like Raman spectroscopy are often too slow for high-throughput applications.

Purpose of the Study:

• To develop a method for rapid, label-free cell characterization using Raman spectroscopy.
• To optimize compressive sensing parameters for enhanced speed and performance.

Main Methods:

• Development of a neural network model to identify optimal parameters for compressive Raman sensing.
• Application of the model to a dataset of Raman spectra from three distinct cell types.

Main Results:

• Measurement time was reduced by two orders of magnitude.
• Achieved up to 90% classification accuracy for different cell types.
• Utilized only five linear combinations of Raman intensities for analysis.

Conclusions:

• The developed neural network model enables high-throughput cell characterization.
• This method significantly accelerates Raman spectroscopy, making it viable for real-time cell product analysis.
• Unlocks the potential of Raman spectroscopy for advancing cell-based therapies.