Machine learning microfluidic based platform: Integration of Lab-on-Chip devices and data analysis algorithms for red blood cell plasticity evaluation in Pyruvate Kinase Disease monitoring

Microfluidics represents a very promising technological solution for conducting massive biological experiments. However, the difficulty of managing the amount of information available often precludes the wide potential offered. Using machine learning, we aim to accelerate microfluidics uptake and lead to quantitative and reliable findings. In this work, we propose complementing microfluidics with machine learning (MLM) approaches to enhance the diagnostic capability of lab-on-chip devices. The introduction of data analysis methodologies within the deep learning framework corroborates the possibility of encoding cell morphology beyond the standard cell appearance. The proposed MLM platform is used in a diagnostic test for blood diseases in murine RBC samples in a dedicated microfluidics device in flow. The lack of plasticity of RBCs in Pyruvate Kinase Disease (PKD) is measured massively by recognizing the shape deformation in RBCs walking in a forest of pillars within the chip. Very high accuracy results, far over 85 %, in recognizing PKD from control RBCs either in simulated and in real experiments demonstrate the effectiveness of the platform.

Automated summary

Microfluidics offers great potential for conducting large-scale biological experiments, but the difficulty of managing available information limits its widespread use. In this study, we propose combining microfluidics with machine learning approaches to enhance the diagnostic capability of lab-on-chip devices. By introducing data analysis methodologies within the deep learning framework, we are able to encode cell morphology beyond standard cell appearance. Using a dedicated microfluidics device, our machine learning platform accurately recognizes Pyruvate Kinase Disease (PKD) in red blood cell samples, achieving over 85% accuracy in simulated and real experiments.