Advancing smartphone-assisted paperbased biosensors with artificial intelligence

Point-of-care testing (POCT) plays a crucial role during large-scale disease outbreaks. It allows non-professionals to conduct early testing on-site and rapidly obtain accurate results. Paper-based biosensors have been demonstrated as a promising tool for detecting disease-relevant biomarkers at POCT thanks to their low cost, ease of operation, and self-driven capillary fluidic flow. In the last decade, many biosensing mechanisms, such as electrochemistry, chemiluminescence, fluorescence, and colorimetric immunoassay, have been developed and realized on paper-based biosensors. Among them, colorimetric paper-based biosensors are particularly suitable for use in low-resource settings, especially when using smartphones as colorimetric reading and analyzing tools (no further complicated equipment is needed). However, Smartphone-assisted colorimetric paper-based biosensors encounter challenges, including poor standardization and consistency, limited accuracy, and limited sensitivity. Machine learning (ML) as a branch of artificial intelligence (AI) has recently been demonstrated as one of the most potent data processing tools, and it can be readily implemented on smartphone platforms. ML can provide novel strategies to overcome the challenges faced by smartphone-assisted colorimetric paper-based biosensors and can also be a pathway for ordinary biosensors to become smart biosensors, which can automatically predict the type or concentration of analytes based on a decision-making system.     This thesis focuses on developing advanced smartphone-assisted paper-based biosensors with AI. This thesis starts with a review of the development of paper-based biosensors, followed by a brief introduction of commonly used readout devices for various biosensing mechanisms. It is pointed out that smartphone readout devices play an important role in POCT. Then, smartphone-assisted paper-based biosensors for various biosensing mechanisms are described in detail. Next, how artificial intelligence can be beneficial for smartphone paper-based biosensors is discussed. It also discusses the process of AI in processing data from paper-based biosensors. Finally, several common AI algorithms are presented. This review provides a foundation for the development of advanced smartphone-assisted paper-based sensors. The first branch of this thesis utilized the most representative paper-based biosensor-lateral flow assay (LFA). By employing ML models, segmentation and detection of LFA test strips were achieved, resulting in a classification accuracy of up to 99% after training. This approach mitigated the impact of different smartphone camera settings on LFA results in image analysis. Additionally, it addressed the current limitation of commercial LFA tests for dry eye disease (DED), which can only perform dichotomous detection. This work not only overcomes the challenge of poor standardization and consistency in smartphone-assisted detection but also facilitates the broader application of LFA in large-scale testing.     However, the detection sensitivity of LFA is insufficient for quantitative analysis. Therefore, more sensitive microfluidic paper-based analytical devices (μPAD) were employed in the second stage of this thesis. The second work applied the ML-assisted smartphone platform for ultra-accurate detection of colorimetric enzyme-linked immunosorbent assays (c-ELISA) using μPADs. Unlike existing smartphone-based μPAD platforms, this platform eliminated the impact of random lighting and improved detection accuracy. A user-friendly smartphone application was also developed to control the whole process, enabling ‘image in, answer out,’ maximizing smartphone convenience. This work successfully addresses the challenges of limited accuracy and sensitivity in smartphone-assisted biosensors.     The third work introduces an ML-assisted offline smartphone platform for early screening of Alzheimer's disease (AD), offering rapid disease detection in low-resource areas. The proposed platform features a simple mechanical rotating structure controlled by a smartphone, enabling fully automated c-ELISA on μPADs. This platform successfully applied sandwich c-ELISA for detecting the β-amyloid peptide 1–42 (Aβ 1–42, a crucial AD biomarker) and demonstrated its efficacy in artificial plasma samples. Moreover, the ML model provides higher accuracy than the traditional method of curve-fitting results. The trained ML model was integrated into the smartphone using the NCNN (Tencent's Neural Network Inference Framework), enabling ML-assisted offline detection. A user-friendly smartphone application was also developed to control the entire process. This work reduces the complexity of manual operation for c-ELISA detection. It also enables offline smartphone detection without requiring large computing devices or cloud data transfer.     In summary, this thesis developed advanced paper-based biosensors from three perspectives. And successfully addresses the challenges faced by smartphone-assisted paper-based biosensors using ML. It further fulfills all the criteria the World Health Organization (WHO) set for POCT devices in resource-poor environments. It also can significantly improve global healthcare by providing low-cost, timely, and accurate disease screening.