At a glance
CDC's AI Accelerator Program (AIX) accelerates the development and scaling of high-impact artificial intelligence (AI) solutions, serving as a repeatable model for piloting and operationalizing AI tools.
Why the effort matters
Accelerating access to analytic and automated solutions to support public health is a core goal of the Public Health Data Strategy (PHDS). One powerful way to accelerate access is by using AI. By automating data analysis, AI has the potential to generate new insights and improve understanding of public health trends and threats.
AI is a vast and evolving field that includes a variety of methodologies, such as computer vision, natural language processing and machine learning. These methods are useful for pattern recognition, prediction and large-scale data analysis.
In public health, AI offers the potential to increase effectiveness, precision and scalability - enabling faster detection of outbreaks, more tailored interventions and improved treatment strategies. AI can also assist in locating novel insights and patterns that human analysts might miss.
The progress
The 2024 PHDS included a milestone to release a plan for how the agency would leverage AI and launch at least three AI use cases, which are solutions to specific challenges. Early outcomes include key scalable solutions.
TowerScout uses computer vision, a subfield of AI, to analyze satellite imagery, automatically detecting cooling towers that may harbor Legionella bacteria. This approach reduces cooling tower identification time per area from four hours to five minutes, a reduction of 98 percent1. It enhances outbreak response by precisely identifying cooling tower locations and helps jurisdictions build cooling tower registries to prevent future outbreaks.
TowerScout gives investigators the data they need in minutes — allowing for faster source identification, fewer infections and lives saved.
Clinical Narratives tackles the challenge of making sense of massive volumes of unstructured clinical notes in electronic health records, which often contain critical information concealed under vast amounts of free text. This AI solution extracts key data, like diagnoses, medications and symptoms. These data help to identify critical markers supporting timely detection of outbreaks and improve surveillance for high-risk populations. It has increased the accuracy and speed of information extraction by 80 percent with respect to a baseline scenario1, helping public health teams act on insights that might otherwise be overlooked.
NewsScape addresses the challenge of monitoring emerging health threats within thousands of daily news stories and open-source reporting. It leverages large language models to enable both case-based and event-based surveillance teams to quickly extract key information from vast amounts of data. This solution can process approximately 8,000 articles a day1 and could increase CDC's ability to monitor potential health threats.
What partners are saying
"Many large Legionnaires' disease outbreaks have been linked to cooling towers. However, it is not feasible to manually review satellite images to identify all cooling towers in a county as large and populous as Los Angeles. TowerScout's image recognition algorithms allowed us to automate this extremely labor-intensive process and identify cooling towers so the county is better prepared to respond to potential future Legionnaires' disease outbreaks." — Dr. Prabhu Gounder, Medical Epidemiologist, Los Angeles County Department of Public Health
- Impact metrics listed, such as time saved, are based on unpublished CDC studies.