A robust smartphone-based deep learning system for accurate quantification of immunochromatographic strips

Purpose This study aims to develop an integrated smartphone-based system for automated and quantitative analysis of immunochromatographic test strips, addressing the need for portable and reliable point-of-care diagnostic tools. The research seeks to improve the robustness and accuracy of strip interpretation under diverse environmental and user-dependent conditions. Design/methodology/approach The proposed system incorporates a sequential deep learning framework, consisting of a convolutional neural network for image quality assessment and strip validity classification, followed by an optimized EfficientNet-B0 regression model for quantitative concentration prediction. The pipeline decouples quality control from quantitative estimation through dedicated model design, enabling reliable performance on smartphone hardware. Findings Experimental results demonstrate that the system achieves classification accuracy across five categories no-target poor-quality invalid-strip negative and valid-positive samples. In addition, the regression module attains high quantitative accuracy with R2 values exceeding 0.98. These results confirm the feasibility of smartphone-based automated strip analysis despite its performance being slightly lower than specialized laboratory instruments. Originality/value This work offers a novel architecture that integrates sequential deep learning modules for quality assessment and quantitative prediction, enhancing robustness while maintaining computational efficiency. The system provides a practical and portable solution for auxiliary diagnosis, home healthcare monitoring and primary healthcare applications, demonstrating strong potential for real-world deployment in telemedicine and point-of-care testing.