Machine learning-based colorimetric determination of glucose in artificial saliva with different reagents using a smartphone coupled mu PAD

Potassium iodide (KI) and 3,3',5,5'-tetramethylbenzidine (TMB) are frequently used as chromogenic agents in mu PADs for glucose determination. Chitosan (Chi) has peroxidase like activity and improves the analytic performance of mu PADs when used in combination with a chromogenic agent. Here, a portable platform incorporating a mu PAD with a smartphone application based on machine learning was developed to quantify glucose concentration in artificial saliva. The detection zones of the mu PAD were modified with three different detection mixtures containing; (i) KI, (ii) KI+Chi and (iii) TMB. After the color change, the images of the mu PADs were taken with four different smartphones under seven different illumination conditions. The images were first processed for feature extraction and then used to train machine learning classifiers, resulting in a more robust and adaptive platform against illumination variation and camera optics. Different machine learning classifiers were tested and the best machine learning classifier for each detection mixture was obtained. Next, a special application called "GlucoSensing" capable of image capture, cropping and processing was developed to make the system more userfriendly. A cloud system was used in the application to communicate with a remote server running machine learning classifiers. Among the three different detection mixtures, the mixture with TMB demonstrated the highest classification accuracy (98.24%) with inter-phone repeatability under versatile illumination.