Lightweight machine learning-assisted dual-line lateral flow immunoassay for simultaneous on-site screening of AFP and CEA
Year: 2026
Type: preprint
Abstract: Smartphone imaging and lightweight machine-learning readouts are extending colloidal gold-based lateral flow immunoassays (LFIAs) from subjective visual tests toward objective and multiplexed point-of-care screening. In this study, a dual-line LFIA integrated with lightweight machine learning was developed to simultaneously determine alpha-fetoprotein (AFP) and carcinoembryonic antigen (CEA), in which two spatially separated test lines generated parallel signals. LFIA images were acquired using a smartphone. Regions corresponding to the control line, two test lines, and their adjacent backgrounds were identified using predefined geometric constraints. Mean RGB values were converted into RGB vector magnitudes, and background-corrected test-line features were obtained by subtracting the corresponding local background signals. The resulting one-dimensional features were analyzed using two lightweight machine-learning algorithms: k-nearest neighbors (KNN) for concentration-interval classification and Gaussian process regression (GPR) for quantitative prediction. Under the optimal conditions, the limits of detection were 4.61 ng mL⁻¹ for AFP and 5.46 ng mL⁻¹ for CEA in human serum. The overall classification accuracy reached 0.91 for both biomarkers. In 32 clinical serum samples, the ROC AUCs were 0.973 for AFP and 0.914 for CEA, increasing to 0.992 when both biomarkers were combined. The dual-line LFIA combined lightweight machine-learning workflow improves the objectivity and analytical utility of the conventional colorimetric LFIAs.
Source: Research Square
Authors: Zhanwei Liang(Conceptualization, Formal analysis, Investigation, Methodology, Visualization, Writing – original draft), Chenxi Wang(Data curation, Investigation, Methodology), Hua Ye(Investigation, Validation), Jincheng Xiong(Resources, Validation), Shang Wang(Data curation, Investigation), Xiaoqing Li(Resources, Supervision, Writing – review & editing), Xinhua Dai(Supervision, Writing – review & editing), Tao Peng(Conceptualization, Funding acquisition, Project administration, Supervision, Writing – review & editing)
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
Awards: G4510012035, G5432701075, G603096694
OpenAlex ID: W7236813905
