Development of an AI image Recognition-based lateral flow immunochromatographic test strip for higenamine detection

Higenamine (HM) is a naturally occurring benzylisoquinoline alkaloid found in plants and is classified as an S3 prohibited substance in the 2020 World Anti-Doping Agency report. To prevent doping violations in competitive sports, it is essential to develop timely and accurate detection methods. In this study, we propose, for the first time, a detection method that combines artificial intelligence (AI)-based image recognition technology with lateral flow immunoassay (LFIA). This method utilizes gold nanoparticles (AuNPs) conjugated with HM-specific antibodies for the rapid detection of HM in urine via LFIA. It integrates AI-based image recognition to quantitatively analyze color intensity of the test strip's detection line. Experimental results demonstrated that the proposed method achieves a limit of detection of 0.49 ng/mL, enabling accurate identification within the concentration range of 2-8 ng/mL. For model training, an Image Classification of HM Test strip dataset was created. To address the limitations of this dataset, namely its small size (n = 304) and limited feature diversity, an advanced solution tailored to these constraints was proposed. The use of Contrast Limited Adaptive Histogram Equalization (CLAHE), which enables finer-grained feature extraction, ensured excellent recognition accuracy of the model in this dataset. During actual urine detection, the model achieved a prediction accuracy of 96.88% on the test set. This approach provides an efficient and reliable technical solution for on-site rapid detection of HM, addressing the critical need for timely monitoring in anti-doping efforts.