ИСТИНА

Smartphone based colorimetric biochemical assay quantification is of increasing interest in food safety analysis. As a result several systems have been developed enabling accurate predictions for specific tests with a smartphone using red green blue (RGB), hue saturation value (HSV) or lightness and chromatic axes A/B (LAB) colour space. However, these systems are test (and often smartphone) specific and have not been used to quantify colour in existing commercial assays. Mainly only one of the three available channels in a colour space is used. For instance, L is commonly used for lateral flow assay (LFA) quantification and has been claimed to work in a superior manner to all other channels mentioned here. However, to the best of our knowledge, this claim has not been validated by robustly comparing (L) performance with other channels. Moreover, digital representation of true colour is strongly simplified by utilizing only one of the three channels of a colour space. Other issues encountered are background illumination variation caused by various light settings and inter-phone variation in the channel values. To tackle these issues a light shielding box as well as camera calibration algorithms have been suggested. Although these solutions are acknowledged to be capable of correcting for large variations, simpler background correction techniques might take care of most of the variations expected during normal use. In the present study, a more universal approach towards smartphone colorimetry was developed. Prediction accuracy of each individual channel of the RGB, HSV and LAB colour spaces was determined for the quantification of colour variation, using pH strip, and colour intensity variation, using filter paper with dropped gold, latex or carbon black nanoparticles. R and B channels were identified as best performing for colour change and colour intensity respectively. A simple background correction was shown to be capable of avoiding variation caused by illumination as long as no direct sunlight was used, thus eliminating the need for a box. Moreover, the same background correction enabled, to a large extent, the elimination of inter-phone (n=6) variation thus permitting the quantification of colour with phone A using a calibration curve constructed on phone B. To validate the universal nature of the optimised system it was successfully used to quantify gluten in buffer (B channel), bovine milk in goat milk (B channel) and pH values in soil extracts (R channel) using commercial assays and various phones (all obtained prediction curves with R2 > 0.9 and mean average errors < 30%). Next a machine learning algorithm and app was developed to allow automated background subtraction and random channel combination of all colour spaces in 2 or 3 channel multidimensional calibration plots. This allowed error reduction and the establishment of a universal, user friendly app able to successfully quantify colour in various commercial LFAs and pH strips.