Incorporation of Discrete Wavelet Transformation, Empirical Mode Decomposition, and Singular Value Decomposition in Microfluidic Fluorescence Flow Cytometry for Noise Reduction

Featuring high miniaturization and portability, microfluidic flow cytometry plays a core role in point-of-care applications of single cell analysis. However, it suffers from a key bottleneck: low signal-to-noise ratios due to miniaturization. Previously, only preliminary denoising algorithms were incorporated for noise reduction. In this work, discrete wavelet transformation (DWT), empirical mode decomposition (EMD), and singular value decomposition (SVD) as representative transform-domain and statistical-model denoising algorithms were incorporated with a home-developed microfluidic fluorescence flow cytometer for noise reduction. The effects of key parameters within these three denoising algorithms on signal-to-noise ratios were investigated systematically, and the effects of single and combined use of these three denoising algorithms on noise reduction were studied in a comparative manner. The optimized denoising algorithm of DWT + SVD reduced the noise standard deviation to approximately one-third of its original value while preserving pulse morphologies. When DWT + SVD was incorporated to process fluorescence signals of microfluidic flow cytometry, distinct pulses emerged in previously noise-obscured waveforms, leading to the location of a new population of low-intensity events. The DWT + SVD denoising algorithm effectively improved the signal-to-noise ratio of microfluidic flow cytometry, providing a new perspective on noise reduction.