Detecting and discriminating between different types of bacteria with a low-cost smartphone based optical device and neural network models

摘要

The food industry such as meat producers and plant product processors have tremendous interest in detecting pathogenicorganisms such as E.coli, Listeria, and Salmonella in very low concentrations down to a single cell. These pathogenicorganisms when they are in the right environment can start multiplying exponentially. For example, E.coli cells candouble every 20 minutes posing a tremendous danger for their growth in over many hours. We have designed an opticaldevice that attaches to a smartphone providing an imaging and processing device that achieves an optical resolution of 1micron. The optics is engineered to reduce aberrations in the system. We also developed a smartphone application thatcan track microbeads and bacteria in the video frames in real time using computer vision algorithms. We extractindividual bacterial image segments in these videos to train neural network models to detect and differentiate differenttypes of bacteria such as E.coli and B.subtilis. These trained models can detect and discriminate E.coli from B.subtiliswith high accuracy of more than 80%. This approach has the potential to train different types of bacteria with amulticlass neural network classifier by training them with images from different genera and species of bacteria. Such aclassifier can detect them in a wild sample containing many types of bacteria with low-cost smartphone optical device.