Event-Preserving Neural Network Denoising for Impedance Flow Cytometry

Impedance flow cytometry (IFC) is a label-free technique for high-throughput electrical characterization of individual cells and other micron-scale particles based on transient impedance changes during passage through microfluidic sensing electrodes. Although lock-in demodulation and low-pass filtering are commonly used for signal conditioning, post-demodulated IFC signals can still contain residual noise, baseline drift, periodic interference, and event-like artifacts that reduce event detectability and distort quantitative features. Here, we present an event-preserving neural network-based denoising framework for post-demodulation IFC signal enhancement. The novelty of this work lies in an IFC-specific event-preserving denoising strategy that combines event-focused sampling and an event-weighted loss to suppress noise while preserving sparse bipolar particle events. A lightweight multilayer perceptron (MLP) was trained and evaluated using paired synthetic noisy and clean IFC signals generated with representative noise and artifact components, and further tested on experimental IFC measurements under challenging noise conditions. Neural network denoising improved event detection, reduced amplitude-estimation error under added white noise, and suppressed experimental baseline and background fluctuations while preserving bipolar event morphology. These results suggest that lightweight neural network denoising can serve as a practical optional enhancement step for noisy post-demodulated IFC signals, potentially supporting more reliable electrical detection and analysis of micro- and submicron-scale particles in impedance-based biosensing applications.