Deep Learning-Assisted Ultrasensitive Detection of Gold Nanoparticles Using Light Microscopy Images Captured by a Cellphone Camera

Gold nanoparticles (AuNPs) exhibit strong light absorption and scattering properties due to localized surface plasmon resonance, making them valuable tools in optical sensing and imaging applications. Direct visual recognition of single AuNPs enables simple and ultrasensitive detection. In this study, we report an approach for the detection and quantification of AuNPs using dark-field scattering light microscopy images captured with a mobile phone camera. Deep learning was incorporated for image analysis to promote ultrasensitive recognition and detection of 120 nm AuNPs with concentrations ranging from 5.3 to 530 fM. Preprocessed images were split into training and testing data to build two deep-learning models, i.e., classification and regression. The classification model achieved perfect precision, recall, and F1 score with a two-image input strategy, while the regression model demonstrated a correlation coefficient of 0.9999 between predicted and actual concentrations. Blind tests of 4 samples at different concentrations confirmed the method's prediction accuracy, with recovery rates of 97-108%. This work presents a simple, easily accessible, and highly sensitive platform for AuNPs detection with potential applications in a wide range of sensing tasks, leveraging the accessibility of mobile phone cameras and the robustness of deep learning techniques.