A deep learning-enabled smartphone platform for rapid and sensitive colorimetric detection of dimethoate pesticide

• Published in: Analytical and bioanalytical chemistry 2023-12, Vol.415 (29-30), p.7127-7138
• Main Authors: Liu, Shuai, Zhao, Jingkai, Wu, Junfeng, Wang, Ling, Hu, Jiandong, Li, Shixin, Zhang, Hao
• Format: Article
• Language: English
• Subjects: Analytical Chemistry; Aptamers; Aptamers, Nucleotide; Artificial neural networks; Biochemistry; Biosensing Techniques - methods; Biosensors; Characterization and Evaluation of Materials; Chemistry; Chemistry and Materials Science; Color; Colorimetry; Colorimetry - methods; Deep Learning; Dimethoate; enzyme-linked immunosorbent assay; Food Science; Gold; Image acquisition; Image enhancement; Laboratory Medicine; Limit of Detection; Machine learning; Metal Nanoparticles; mobile telephones; Monitoring/Environmental Analysis; Neural networks; oligonucleotides; Pesticides; regression analysis; Regression coefficients; Research Paper; Smartphone; Smartphones; Sodium chloride
• ISSN: 1618-2642; 1618-2650; 1618-2650
• DOI: 10.1007/s00216-023-04978-z

A novel deep learning-enabled smartphone platform is developed to assist a colorimetric aptamer biosensor for fast and highly sensitive detection of dimethoate. The colorimetric determination of dimethoate is based on the specific binding of dimethoate and aptamer, which leads to the aggregation of AuNPs in high-concentration NaCl solution, resulting in an obvious color change from red to blue. This color change provides sufficient data for self-learning enabled by a convolutional neural network (CNN) model, which is established to predict dimethoate concentration based on images acquired from a smartphone. To enhance user-friendliness for non-experts, the CNN model is then embedded into a smartphone app, enabling offline detection of dimethoate pesticide in real environments within just 15 min using a pre-configured colorimetric probe. The developed platform exhibits superior performance, achieving a regression coefficient of 0.9992 in the concentration range of 0–10 μM. Moreover, the app’s performance is found to be consistent with the ELISA kit. These remarkable findings demonstrate the potential of combining colorimetric biosensors with smartphone-based deep learning methods for the development of portable and affordable tools for pesticide detection. Graphical Abstract