Microfluidic E-tongue to diagnose bovine mastitis with milk samples using Machine learning with Decision Tree models

Abstract

We report an electronic tongue based on impedance spectroscopy to detect Staphylococcus aureus and diagnose bovine mastitis in milk samples. This was achieved with optimized sensing units made with layer-by-layer films and by treating the capacitance data with machine learning algorithms employing decision trees models. These films were made with chitosan, chondroitin sulfate, sericin and gold nanoparticles /sericin, whose molecular-level interaction with S.aureus depended on the architecture according to PM-IRRAS measurements. The limit of detection in blank milk varied from 3.41 to 2.01 CFU/mL depending on the sensing unit. This sensitivity was complemented with the selectivity provided by combining the electrical responses of the four sensing units. Indeed, with machine learning it was possible to determine multidimensional calibration spaces (MCS) that could generate rules to explain how the milk samples could be discriminated. With a 7-dimension MCS, distinct S.aureus concentrations could be distinguished from possible interferents with a 100 % accuracy. In crude milk samples, 94 % accuracy was obtained with a 6-dimension MCS in multiclass classification for milk from different udders of a mastitis infected cow, including samples diluted 50-fold, in addition to milk from an infected cow treated with Bronopol and from a healthy cow. It is significant that in a ternary classification with these crude milk samples, a 2-dimension MCS could distinguish between milk from an infected cow, treated with Bronopol and from a healthy cow with 100 % accuracy. The combination of electronic tongues and machine learning – as in this proof-of-concept study - is promising for diagnosis of mastitis at a low cost.