Machine learning assisted multi-level differentiation of blood cells using multi-frequency electro-mechanical impedance cytometry

We report a multi-frequency microfluidic impedance cytometer with multimodal measurement of electrical and mechanical cell properties. Using the large feature set from the multimodality, we implemented machine learning assisted differentiation of cells. We have demonstrated a constriction-equipped channel to study the mechanical cell properties, and integrated multiple microelectrodes that capture electric field information pertaining to various stages of cell transit. Our system works over a wide frequency range from 800 kHz to 8 MHz. We developed a machine learning workflow to classify cells using our multi-modal data. Using this platform to show blood cell differentiation, our system shows minimum F1 score to be 0.92, minimizing both false negatives and false positives, which is important for medical diagnosis. Furthermore, we also identified the presence of two populations in the mixed sample of neutrophils and PBMCs. Our workflow allows classification of cells in a new sample along with identification of top 5 distinguishing parameters out of 70+ features extracted from the raw impedance signal.