Nanozyme-based colorimetric sensor arrays coupling with smartphone for discrimination and "segmentation-extraction-regression" deep learning assisted quantification of flavonoids

• Main Authors: Qin, Yuelian, Zhong, Xinyu, Liang, Caihong, Liang, Zhenwu, Nong, Yunyuan, Deng, Lijun, Guo, Yue, Li, Jinfeng, Zhang, Meiling, Tang, Siqi, Wei, Liuyan, Yang, Ying, Liang, Yonghong, Wu, Jinxia, Lam, Yeng Ming, Su, Zhiheng
• Other Authors: School of Materials Science and Engineering
• Format: Article
• Language: English
• Published: 2024
• Subjects: Engineering; Nanozyme; Colorimetric sensor array
• Online Access: https://hdl.handle.net/10356/180745
• Institution: Nanyang Technological University

Achieving rapid, cost effective, and intelligent identification and quantification of flavonoids is challenging. For fast and uncomplicated flavonoid determination, a sensing platform of smartphone-coupled colorimetric sensor arrays (electronic noses) was developed, relying on the differential competitive inhibition of hesperidin, nobiletin, and tangeretin on the oxidation reactions of nanozymes with a 3,3',5,5'-tetramethylbenzidine substrate. First, density functional theory calculations predicted the enhanced peroxidase-like activities of CeO2 nanozymes after doping with Mn, Co, and Fe, which was then confirmed by experiments. The self-designed mobile application, Quick Viewer, enabled a rapid evaluation of the red, green, and blue values of colorimetric images using a multi-hole parallel acquisition strategy. The sensor array based on three channels of CeMn, CeFe, and CeCo was able to discriminate between different flavonoids from various categories, concentrations, mixtures, and the various storage durations of flavonoid-rich Citri Reticulatae Pericarpium through a linear discriminant analysis. Furthermore, the integration of a "segmentation-extraction-regression" deep learning algorithm enabled single-hole images to be obtained by segmenting from a 3 × 4 sensing array to augment the featured information of array images. The MobileNetV3-small neural network was trained on 37,488 single-well images and achieved an excellent predictive capability for flavonoid concentrations (R2 = 0.97). Finally, MobileNetV3-small was integrated into a smartphone as an application (Intelligent Analysis Master), to achieve the one-click output of three concentrations. This study developed an innovative approach for the qualitative and simultaneous multi-ingredient quantitative analysis of flavonoids.