AI-Assisted Analysis to HIV Lateral Flow Test with a Hybrid Model Running on a Distributed Diagnosis System

As a rapid, convenient, and low-cost detection method in point-of-care (POC) diagnosis, lateral flow immunoassay (LFIA) based on chromatography is widely used in different fields, for example, rapid screening test to HIV infection. For LFIA with colloidal gold labeling, traditionally, the test result can be observed with eyes, or identified with image processing algorithm. In some special cases, for example, inhomogeneous chromatography, analytes with low-concentration, and contamination of lateral flow strip with red blood cells or other material, it is difficult to accurately identify the test result with traditional methods. To perform accurate, convenient and easy-to-operate HIV test in POC diagnosis, a distributed diagnosis system, which relies on a handheld device for field test, and as well as a remote server for analysis of detection result, is developed based on the concept of cloud platform. To achieve more sensitive and more specific detection with lateral flow strip, a hybrid model (“traditional model + deep learning model”) is adopted to effectively identify the test area (T-line) of the lateral flow strip. The traditional model running on the handheld device is used to identify significant positive test, and meanwhile, the deep learning model running on the server is used to identify low positive or unexpected test. Principal component analysis (PCA) is used to further improve the performance of the deep learning model when dealing with whole blood test. It has been demonstrated that, with the hybrid model, the accuracy of HIV detection with LFIA can be significantly improved especially when dealing with low positive tests or some special cases, which is beneficial to optimize the performance of HIV screening test in POC diagnosis.