A Machine Learning-Driven Cyclic Optimizing Strategy for the Construction of Paper-Based Microfluidic Devices in the Early Diagnosis of Periodontitis.

The lack of effective optimization strategies hinders the optimal performance of paper-based microfluidic analytical devices (μPADs). In this work, a Machine Learning-driven Computer vision-BP Neural Networks-Genetic Algorithm-based Cyclic Optimizing Strategy (CNGCOS) has been explored to assist in the parameter optimization and engineering of the μPADs. With dual-signal output of color intensity (CI) and colorimetric distance (CD), the optimized μPADs can serve for rapid point-of-care detection of salivary hemoglobin (Hb), an early biomarker for the diagnosis of periodontitis. Moreover, the CNGCOS-assisted μPADs demonstrates high accuracy and superior sensitivity, with an R2 value of 0.998 and a detection limit as low as 1.57 μg/mL for CI output, and an R2 value of 0.992 with a detection limit of 3 μg/mL for CD output. Furthermore, the constructed CNGCOS-assisted μPADs have been applied for the analysis of clinical saliva samples for early diagnosis of periodontitis. Successful detection in 103 clinical cases further validates the exceptional performance and accuracy of the CNGCOS-assisted μPADs. Therefore, the explored CNGCOS has great potential for the optimization of engineering devices for early diagnosis and treatment of diseases.