Food Protection Trends - November/December 2024 - 404

data (72). Using L. monocytogenes in milk as an example,
Njage at al. (47) demonstrated how machine learning
(ML) can be used to predict stress phenotype components
for new unknown pathogen strains given their whole
genome sequencing (WGS) data, which could improve risk
assessments for foodborne pathogens. Agent-based models
(ABMs) developed to recreate a specific food facility, have
provided facility-specific " personalized " food safety decision
support for the food production environments (76). A study
by Nogales at el. (48) demonstrated the utility of neural
models in optimizing the number of food safety inspections
using data from the Rapid Alert System for Food and Feed
(RASFF), which facilitates the exchange of information
about health threats in European countries. Importantly,
a number of peer-reviewed studies have developed and
detailed specific tools that can be applied to CEA, including
simulation-based and ML tools that can be used to predict
food safety risks and facilitate improved control. For
example, a series of papers (3, 4, 65, 76) has detailed how
ABMs can be developed and used to facilitate improved
control of environmentally transmitted foodborne pathogens.
Also, a simulation model was developed to assess the
listeriosis risk associated with a contaminated production lot
of frozen vegetables (77) as an example of a decision-support
tool for food safety and business management that can also
be applied to CEA. Similarly, a number of different decision
tree and ML-based methods have been applied to predict
times and locations with an increased risk of pathogen
contamination in fields and water sources (62, 63, 73, 74);
these approaches could also be adapted to CEA, even though
input data would obviously be very different.
One key opportunity for digital and AI tools in CEA will
be to integrate AI-based and digital food safety prediction
and management tools into the overall digital infrastructure
for CEA (i.e., systems that manage temperature, lighting,
etc.). The ultimate goals of these efforts would be to (i)
reduce food safety risks and manage them more effectively,
(ii) minimize food safety-associated enterprise risks (e.g.,
by developing and implementing systems that can provide
for validated " clean breaks " between lots), and possibly to
(iii) provide enhanced transparency of food safety efforts
to customers and possibly regulatory agencies. While these
efforts will require (i) robust data acquisition systems and
(ii) custom-tailored AI and digital tools, as detailed in the
section above, existing tools that represent a starting point
for these efforts already exist. Ultimately, proactive food
safety systems for CEA could benefit from the development
of comprehensive " digital twins " (20), which would be
individualized to each facility and be driven by real-time data
to predict food safety risks. These digital twins could then be
used to adjust food safety measures (e.g., testing frequencies,
sanitation procedures, water treatment) to better control
food safety risks and better manage situations with increased
risk of pathogen contamination. Additionally, these models
404 Food Protection Trends November/December
can also be complemented by various imaging or spectral
based AI models to identify hot spots for organic build-up
on diverse surfaces and Zones in CEA facilities (25). A
connection between AI models and microbial testing results
can further aid in enhancing the relevance of these models
for the assessment of food safety risks. This would allow
CEA facilities to understand how different environmental
and structural factors interact and affect the probability of
foodborne pathogen (e.g., Salmonella and L. monocytogenes)
contamination in different CEA facilities, and potentially
reduce the cost of microbial sampling. Finally, these digital
twins could be used to concurrently manage food safety as
well as other outcomes such as quality, productivity etc.,
which could be used to optimize trade-offs between food
safety, productivity, and quality. An example of such a
trade-off is related to the negative impacts of high levels of
chlorine that may be used, at the cost of quality, for water
treatment and management of food safety risks. Additionally,
combining models to manage food safety as well as quality
and productivity will make it a more attractive product for
the industry to implement.
More targeted food safety-related applications of AI
in CEA could address the unique food safety challenges
posed. Primary inputs (e.g., substrate, water, seeds) present
a pathway of contamination in CEA, so the development
of AI approaches based on imaging and spectral analysis of
primary inputs used in CEA could enable validation of the
quality and authenticity of them. Furthermore, these tools
can also enable the evaluation of these primary inputs during
storage and handling. Developing ML/AI tools based on
predictive relationships between the imaging or spectral
features of these primary inputs and microbial analysis could
provide a potential surrogate marker for assessing microbial
contamination (52).
Water, as a critical primary input and route of widespread
contamination, is an important point for microbial
monitoring (57). Microbial contamination of water can be
addressed by developing a robust analysis of water quality
and treatment of water. AI approaches can aid in automating
the analysis of water quality, and develop predictive
models of water quality for different seasons based on the
geographical location and local water sources used for CEA
operations (30, 49, 68). In addition, AI approaches could
also be developed to improve and evaluate the efficacy of
water treatment technologies.
CHALLENGES FOR AI TOOL APPLICATIONS TO
FOOD SAFETY IN CEA
While there is enthusiasm for the use of digital and AI
tools to improve food safety (e.g., the US FDA's " Food Safety
Initiative " ) (37), there are limited examples of successful
practical applications of these tools to CEA food safety.
Due to the relative newness of the CEA sector, a specific
challenge faced by many firms is a lack of foundational food
Food Protection Trends - November/December 2024

Table of Contents for the Digital Edition of Food Protection Trends - November/December 2024
