Mars Science & Diagnostics uses Azure AI to deliver better animal health outcomes

Pets can’t speak, so pet owners and veterinary professionals rely on diagnostic tools to help diagnose and monitor health and wellness.

As a division within Mars Petcare, underneath the global Mars brand umbrella, Mars Science & Diagnostics is focused on providing more effective, efficient diagnostic decision-making options for better animal health outcomes.​

“We believe that pets make the world a better place. Our team is dedicated to enabling pets to live their healthiest lives through science, data, and technology,” says Jerry Martin, Vice President of Research & Development at Mars Science & Diagnostics. 

Interpreting data quickly and correctly is key to diagnosing a medical condition, and this can be more challenging with unstructured data such as a scan. 

Mark Parkinson, Senior Director of AI Development at Mars Science & Diagnostics, says, “Veterinary radiologists are key partners in helping veterinary professionals understand diagnostic results, and their expertise is in high demand. We’re looking for ways to support veterinary professionals with tools for diagnosing and monitoring for better health outcomes.”

To solve these challenges, the teams sought a solution to unlock technological potential. 

Getting accurate answers to pet parents faster

Mars has a longstanding relationship with Microsoft, working to become more cloud native and reap the benefits of AI-driven capabilities. Having started with a self-managed database, deploying models on-premises with single containers, Mars knew it needed an intuitive, scalable system. So, the company turned to the Azure AI model catalog and chose Mistral for its models as a service (MaaS) offering, making it easier to build generative AI apps and accelerate the creation of its own RapidRead solution.

“The Azure AI catalog provides access to a wide range of pre-built models such as Mistral to help restructure data and enhance our accuracy. We know it’s accurate because nothing goes into our production systems without being validated and signed off by radiologists,” says Parkinson.

The teams brought together experts from different backgrounds and occupations to develop RapidRead. This involved radiologists, veterinarians, and clinical leaders to identify the keywords needed to arrive at various diagnoses. With Mistral, Mars then worked to fully understand the words these experts provided, using AI to encompass alternative spellings and associated verbiage.

“Working together toward a goal, seeing the alerts for serious findings emerge, and routing those directly to a radiologist—it all makes this project incredibly rewarding,” says Michael Fitzke, AI Senior Director at Mars Science & Diagnostics.

Also employing Azure Machine Learning, Mars teams can focus all their efforts on validation and getting the best results. Within a dedicated workspace, Mars is conducting large-scale training, ingesting more data more quickly from millions of images, which often have millions of accompanying reports. Mars is using Machine Learning to scale up, filtering data and extracting key information at speed. 

Parkinson adds, “The Azure Machine Learning platform significantly enhances large-scale model training, so we could run incredibly large X-ray models. We believe these may be the largest in the veterinary industry.”

Supporting human experts in their work

The teams are also harnessing advanced algorithms to improve RapidRead’s diagnostic accuracy. Through the analysis of radiological images, these algorithms can identify abnormalities that might be missed by a human. 

“We’re using Azure to build more and better AI models for RapidRead to help support the high demand on veterinary radiologists. Additionally, the benefit of scale helps us boost the accuracy of our AI models and expand our operations,” says Martin.

AI capabilities have the potential to unlock new clinical workflows, by assisting with image annotation, report generation, and data management, so radiologists can focus their energies on more complex cases. 

Parkinson says, “By ensuring radiologists have been an intrinsic part of our development and production, we made RapidRead to be complementary to the job of a radiologist.”

Harnessing the full power of Azure solutions

The team’s implementation of a comprehensive range of Azure products, including AI, databases, and app services, means the company can continue refining RapidRead.

Specifically for its work exploring pet dental care, the teams are using Durable Functions, employing the tool to easily run activities in parallel, decreasing the overall time it takes workflows to complete. Using Durable Functions, the teams are creating complex workflows in a serverless environment, focusing instead on business logic without worrying about managing infrastructure.

Nicole McNally, AI Scientist at Mars Science & Diagnostics, says, “Working with Azure makes everything we’re doing quicker. Instead of running one experiment, we can run many, training the model with much bigger datasets.”

With Azure Kubernetes Service (AKS), the teams can choose the best underlying compute for each workload and automatically scale for resource efficiency, freeing engineers from having to manually configure infrastructure. Also, by hosting its endpoints on AKS, the teams are realizing scaling benefits from Azure Service Bus.

The RapidRead tool employs Azure Cosmos DB as its main database, which logs radiologist feedback to further train the model, bringing more findings into the system. Mars chose Azure Cosmos DB for its ability to handle unstructured data, high scalability, low latency, and quick results. Azure Cosmos DB works seamlessly with other Azure services, managing workloads efficiently across multiple regions.

The future of AI in veterinary science

Mars is changing the traditional process of diagnosing a pet in medical need by using Azure. Pets with critical, undiagnosed conditions can receive the care they urgently require, and veterinarians are better able to make informed decisions and get treatment plans underway.

Martin adds, "Clinics are now getting results in just minutes, meaning that critical cases can be flagged without delay, which can support better pet health outcomes. For Mars, that is the best metric by which we can measure our success."