Droplet microfluidics has enabled investigators to generate and manipulate tiny droplets with remarkable precision in a high-throughput manner. However, efficient analysis of tens of thousands of droplets remains a challenge. The integration of artificial intelligence (AI) with droplet microfluidics enables advancements in high-throughput screening and analysis. One of the most promising applications of AI in droplet microfluidics is automated droplet screening and sorting. Traditional manual sorting methods are labor-intensive and time-consuming. AI-driven systems can rapidly analyze droplets based on various characteristics, such as size, composition, and fluorescence, allowing for high-throughput screening of droplets, which is invaluable in applications like drug discovery, cell analysis, and DNA sequencing. In this review, Tsai et al. discusses the integration of microfluidics with automated machine learning for various applications including cell analysis, drug development, and clinical diagnostics. Through AI-driven analysis, optimization, and automated sorting, droplet microfluidics has reached new heights in terms of efficiency and accuracy.
In the field of cell analysis, AI integration with microfluidics has expanded data processing capabilities, enabling prediction and pattern recognition for phenotype identification. Cell counting and classification have been improved by using machine learning algorithms on various data types, including optical images, electrical data, and combinations of both. AI-powered flow cytometry has facilitated high-speed cell counting, classification, and tracking. Deep learning models have been employed to differentiate between different cell types and predict cell properties. For example, Wang et al. employed a deep learning-based segmentation with Bayesian filtering to accurately track the migration of cells within a 3D angiogenic vessel, achieving an impressive 86.4% accuracy. By leveraging high-throughput microfluidic platforms and AI-powered tracking, researchers can now study cell migration guided by physical or chemical gradients without labored manual efforts. This AI-driven cell tracking approach not only reduces the risk of human bias but also enables the quantification of cell migration and growth kinetics with greater precision. Moreover, the ability to accurately track cells through multiple divisions allows for the construction of lineage trees, providing valuable insights into the population dynamics of cells throughout the experimental period. Ulicna et al. successfully reconstructed cellular lineages in extensive timelapse data. This method revealed cell cycling heterogeneity and linked cycles between cells of similar generations. The integration of machine learning with microfluidics has paved the way for more sophisticated cell analysis and offers new insights into cellular behavior and disease mechanisms.
Integration of AI with microfluidics has also proven to be a powerful combination for monitoring human physiological status and detecting biomarkers. Conventional quantification methods using reference interpolation are standardized, but AI presents unique opportunities to quantify and detect anomalies in long-term clinical data through pattern recognition. AI-enhanced microscopic image analysis of cells and tissues has been used for disease diagnosis in clinical settings, such as identifying and monitoring sickle cell anemia and hereditary hemolytic anemia. AI algorithms have also shown promise in diagnosing sepsis and accelerating cancer cell classification and quantification in liquid biopsy and digital pathology applications. For instance, McRae et al. introduced a compact programmable bionanochip platform (pBNC) capable of performing multiplex immunoassays against prostate cancer, ovarian cancer, acute myocardial infarction, and drugs of abuse on disposable cartridges, making it suitable for point-of-care testing applications. The collected data were then sent to machine learning algorithms for further diagnosis based on previously trained data. Additionally, machine learning has facilitated the resolution of optical-based immunoassay readouts. Song et al. presented a digital-enzyme-linked immunosorbent assay microarray for the simultaneous quantification of cytokines in cancer therapy patients. The microwell-based microarray and automated AI image analysis for biomarker quantification reduced the turnaround time to within 40 minutes, enabling rapid and efficient analysis.
In summary, AI can revolutionize the field by enabling intelligent control systems that adapt to real-time changes. AI-based image analysis automates object detection and tracking in microfluidic systems, providing valuable insights and predictions of their dynamics. This integration could lead to advanced lab-on-a-chip diagnostic devices, efficient drug delivery systems, and versatile monitoring platforms, benefiting fields like quantitative biology and personalized medicine. The advantages of AI integration include automation, efficiency, and quantitative analysis, as well as cost-effectiveness due to microfluidics' low reagent consumption. AI's ability to learn from data, even without prior annotation, makes it versatile for complex data analysis tasks. More importantly, it reduces human bias. However, further investigation is needed to ensure robust and user-friendly AI algorithms to address computational complexity and data size issues. Collaboration for data sharing and standardized protocols can help overcome these challenges. Overall, integrating microfluidic platforms with machine learning holds great promise for healthcare breakthroughs, but careful validation and risk assessment are essential.
- Tsai HF, Podder S, Chen PY. Microsystem Advances through Integration with Artificial Intelligence. Micromachines (Basel). 2023 Apr 8;14(4):826. doi: 10.3390/mi14040826. PMID: 37421059; PMCID: PMC10141994.
- Wang J, Liang K, Zhang N, Yao H, Ho TY, Sun L. Automated calibration of 3D-printed microfluidic devices based on computer vision. Biomicrofluidics. 2021 Mar 10;15(2):024102. doi: 10.1063/5.0037274.
- Ulicna K, Vallardi G, Charras G, Lowe AR. Automated Deep Lineage Tree Analysis Using a Bayesian Single Cell Tracking Approach. Front. Comput. Sci. 2021;3:734559. doi: 10.3389/fcomp.2021.734559.
- McRae MP, Simmons GW, Wong J, Shadfan B, Gopalkrishnan S, Christodoulides N, McDevitt JT. Programmable bio-nano-chip system: a flexible point-of-care platform for bioscience and clinical measurements. Lab Chip. 2015 Oct 21;15(20):4020-31. doi: 10.1039/c5lc00636h. Epub 2015 Aug 26. PMID: 26308851; PMCID: PMC4589532.