Machine learning pipeline for microparticle size classification in self-mixing interferometric signals for flow cytometry

Self-mixing interferometry (SMI) is an emerging optical sensing technique for detecting and classifying microparticles in non-contact and label-free flowmetry applications. High precision and reliability are essential for its integration into medical diagnostics, such as blood analysis, and quality control in chemical manufacturing processes. While theoretical models describe SMI-induced signal modulations caused by particle passage, challenges persist due to signal noise, variability, and interpretability under experimental conditions. This study enhances SMI-based particle size classification by integrating machine learning (ML) models to improve feature extraction and classification accuracy. Three ML pipelines are evaluated, achieving 98% classification accuracy in distinguishing particles of different sizes (2, 4, and 10 µm). The high classification accuracy demonstrates the scalability of our approach, ensuring its applicability across diverse particle analysis scenarios.