Nondestructive foodborne pathogen detection using a colorimetric sensor enabled by machine learning and non-toxic dyes

Introduction:

Current foodborne pathogen detection methods are unsuitable for use in the food supply chain post-packaging, but a non-destructive sensor can address this gap in food safety infrastructure by performing in-packaging continuous testing for foodborne pathogens.

Purpose:

This project aimed to develop an AI-enabled sensor for non-destructive sensing of foodborne pathogens using non-toxic dyes’ chromogenic reactivity to volatile organic compounds (VOCs).

Method:

Ground beef (48-60 replicates per treatment) was inoculated with E. coli O157:H7 (3 log CFU/g) or sterile PBS, and then a paper with the non-toxic dyes was exposed to the samples and incubated at 4°C and 25°C for up to 7 days. The color data and ground truth (inoculation scenario, incubation temperature, and incubation time) were used to build neural networks in Python with 5-fold validation, with the color data being used to detect the pathogen.

Results:

The color differences between the control and treatment groups were generally most apparent with longer incubation time and greater incubation temperature. The color change patterns could be interpreted through neural networks for pathogen detection, with detection accuracy as high as 93%.

Significance:

This sensor’s novelty lies in its non-destructive and non-toxic traits, which make it especially fit for smart packaging that can fill the current gaps in foodborne pathogen sensing post-processing. There is great potential for this sensor design to be expanded to detect more pathogens in a wider range of food products, reducing foodborne illnesses and recalls.