Integration of artificial intelligence and microfluidics for disease diagnosis: a systematic review

Background and aims: Microfluidics has been developed explosively in the last two decades dueto its advantages, such as enabling controllability over multiple processes, low consumption ofreagents, and cost-effectiveness. Furthermore, microfluidic systems can provide new technologyfor diagnosing various diseases by analyzing various biomarkers in plasma, serum, and wholeblood. The advantage of these systems is the no invasiveness; they do not require a tissue biopsyto diagnose diseases like cancer. Utilizing microfluidic systems in multiple applications leads tothe generation of large datasets. Traditional tools to analyze this data are not efficient. Thus, newtools are required to analyze this big data. Artificial intelligence could serve proper techniqueswith advantages for this purpose. Artificial intelligence can classify and predict diseases usingbig data from microfluidic systems. Combining microfluidic systems with artificial intelligencetools could make cost and time-saving systems. This study is devoted to the recent developmentsin a microfluidic device coupled with artificial intelligence technology for diagnosing variousdiseases.Method: An electronic search was performed in three databases, including PubMed, Scopus,and Web of Science. In the search strategy, words relevant to “microfluidics” and “artificial intelligence”were searched at the title. Also, words related to “diagnosis” were searched at the titleand abstract levels. The search was not restricted to publications of a specific time range. Theinclusion criteria included the relevance to the search strategy without a time limit and the exclusioncriteria included review, conference, non-English, and duplicated papers. After obtaining thepapers, the articles were investigated from the point of view of the determined topic to identifythe methods and diseases diagnosed by these systems. Type of cell or tissue, type of microfluidicdevice, dataset size, data set type, imaging modality, AI method, and performance were investigatedas common criteria to classify the included papers.Results: Our search identified ۶۹ records from PubMed, ۱۶۳ records from Scopus, and ۹۸ recordsfrom ISI. ۱۷۷ duplicated records were excluded. After applying the inclusion and exclusioncriteria ۳۳ papers were included. Microfluidics can significantly benefit from artificial intelligence.Artificial intelligence accelerates the analysis of massive data gained from microfluidics.Artificial intelligence tools, including deep learning (convolutional neural networks, generativeadversarial networks), data mining (Gradient Boosting Machine, Random Forest, Naive Bayes),and microfluidics, could apply in different diagnosis areas by cell classification and cell isolation.Cell classification is performed by flow cytometry to classify various types of cells, which is verycrucial for the prevention and diagnosis of different diseases. Artificial intelligence provides greataccuracy in cell classification, which was previously unachievable due to the difference betweencells in size, shape, and some external effects. In addition, cell isolation is a critical process tounderstanding the concept of precision medicine.Conclusion: Integrating artificial intelligence and microfluidics could play an important role inbiological and biomedical applications. Next-generation monitoring systems may be developedby integrating microfluidic devices with artificial intelligence. Contrary to the challenges of thesesystems, such as the selection of the appropriate model and data heterogeneity, many fields ofbiotechnology, like point-of-care systems, personalized medicine, diagnostics, and treatment of diseases, can benefit from the combination of artificial intelligence and microfluidic, which for allthese areas, more extensive studies are required.