Korea University: Development of a platform using an AI-assisted smartphone-based assay for early-stage virus infection diagnosis

The research team, led by Professor Yoon Dae-seong from the School of Biomedical Engineering at the College of Health Science of Korea University (KU), has collaborated with the teams of Professor Lee Jeong-hoon (Research Supervisor) from Kwangwoon University, Professor Lee Ki-baek from Kwangwoon University, and Professor Cho Sung-yeon from Seoul St. Mary’s Hospital, to achieve a significant milestone. They have successfully developed AI-assisted smartphone-based lateral flow assay (LFA) diagnostics. This groundbreaking technology enables precise and sensitive decision making using only a smartphone and an LFA test kit.



Lee Seung-min, an Integrated Program student from KU’s School of Biomedical Engineering made a significant contribution to this study as first author and achieved the remarkable feat of consecutively publishing papers in Nature’s sister journal.

COVID-19 LFA kits play a critical role in delivering on-site decisions during the era of pandemics and endemics, with millions of uses every day. Improving the accuracy and sensitivity of these diagnostics can have a significant impact on individual patient treatments and public health. In particular, early diagnosis and screening are crucial before symptoms manifest and for asymptomatic cases. Detecting infections at this stage is crucial for minimizing virus transmission. While PCR has become the standard diagnostic method due to its high sensitivity, it comes with the drawbacks of high diagnostic cost and the potential for unnecessary isolation since it can yield positive results even during the recovery period when virus spread is minimal. On the other hand, while LFA kits offer affordability and rapid results, their application has the drawback of significantly lower sensitivity. Specifically, LFA kits exhibit less than 50% sensitivity for patients in early stages with low viral loads and they face significant challenges in accurately identifying infected patients.

In an effort to address the limitations of LFA test kits, the research team developed a smartphone-based diagnostic technology utilizing a deep learning algorithm to aid in positive/negative decisions. Subsequently, the team conducted a blind test involving 1,500 participants to evaluate the sensitivity of AI-assisted diagnostics. The results revealed that achieving accurate positive/negative determinations with 100% sensitivity was possible, compared to the average sensitivity of 72% achieved by visual inspection in the general population. Particularly significantly, with asymptomatic or early-stage diagnoses, the sensitivity increased substantially from 51% in the general population to 91% when AI was applied. This demonstrates the potential for early-stage diagnosis. Additionally, app-based tests with eight different LFA models available on the market exhibited an average sensitivity of 94.8% and specificity of 90.9%, confirming the versatility of the approach..

This smartphone-implemented AI-assisted technology offers several advantages. First, it allows for early diagnoses with high sensitivity and accuracy. Second, it has the potential for continuous improvements in accuracy through data acquisition and learning. It can also detect concentrations that may not be visible to the human eye, thereby reducing inter-individual variation. Furthermore, it facilitates the digitization and real-time integration of data and enables the assessment of disease progression, improvement, and isolation status through continuous monitoring. Moreover, it can be applied to the diagnosis of emerging virus variants. Lastly, it is capable of diagnosing infections regardless of the specific LFA model or smartphone being used.

This technology has been transferred to CALTH, who will now focus on commercializing it. This process involves optimizing the app and algorithm, as well as obtaining approvals and certifications from regulatory bodies such as the U.S. and Korean Food and Drug Administrations. This research was supported by the Bio & Medical Technology Development Program of the National Research Foundation of Korea and published in Nature Communications, a prestigious journal published by the Nature Portfolio with an impact factor of 17.69.