DHS Office of Health Security – AI Use Cases

Use Case Name: MiX Phenotyping 

Use Case ID: DHS-2418 

Use Case Summary: The goal for this proof-of-concept project is to automate creation of health security insights from DHS electronic medical record data. Clinical notes taken during clinical medical provision will be parsed into key terms and phrases, organized into machine-readable data, and integrated with other data such as geographic location of relevant DHS facilities. AI/ML is used to automate this data processing and quickly identify new DHS clinical patterns. These patterns can provide key health security insights, such as emerging disease detection (anomaly detection), and provide decision support to medical and public health community partners. 

The main AI system outputs will include anomaly detection for emerging trends in clinical record patterns. AI/ML outputs will identify these clinical patterns by clustering, classification, and topic modeling. Finally, these clinical patterns will be presented as linear reference models that are simple enough to be interpreted and guided by human clinicians using human-in-the-loop collaboration. 

Phenotyping will use only de-identified medical data that is publicly available or such de-identified data in DHS care and custody whose use is authorized under a specific Information Exchange Agreement.    

Use Case Topic Areas: Health & Medical 

Deployment Status: Pre-deployment (Initiation) 

Safety- and/or rights-impacting? No, use case is too new to fully assess impacts; will be reassessed before end of initiation stage. 

Face Recognition/Face Capture (FR/FC)? No 

Use Case Name: MiX Indicators 

Use Case ID: (DHS-2419) 

Use Case Summary: The goal for this proof-of-concept project is to quickly detect unusual trends in online news stories related to health security threats. Open-source media are pulled from the internet, and key search terms such as pathogen or chemical names are identified and tallied across articles. AI/ML makes it possible to read, organize, and predict normal trends in news stories, making it possible to quickly detect any unusual trends (anomalies). This capability helps us to better prepare for, respond to, and protect from health security threats. 

The main AI system output will be anomaly detection, which represents two elements: 1) AI/ML predictions for usual trends in news story key terms, and 2) the actual daily number of mentions for key terms in the news. When the actual number of mentions for key terms exceeds the modeled predictions, these instances will be detected as anomalies. 

Use Case Topic Areas: Health & Medical 

Deployment Status: Pre-deployment (Initiation) 

Safety- and/or rights-impacting? No, use case is too new to fully assess impacts; will be reassessed before end of initiation stage. 

Face Recognition/Face Capture (FR/FC)? No 

Use Case Name: MiX MedINT (Medical Intelligence Dashboard and Canvas) 

Use Case ID: DHS-2420 

Use Case Summary: The goal for this proof-of-concept project is to create wholistic visuals of a wide variety of health security data. These data will be pulled from online open sources to broadly include, for example, news on societal conflicts, scientific literature, and reports on weather and aggregated (not individual level) information on disease for humans and wildlife. Visuals will be created using two different approaches. One approach (MedINT Dashboard) is to view global mapping of the health security events, where each event is tagged on the map to its reported location. The map can be modified to view different health security layers and zoomed in/out to locations of interest. The second approach (MedINT Canvas) is to build a data visualization of interest by selecting topics, such as a disease, symptom, and location, and then observing if and how these topics connect through recent health security events. AI/ML will make this project possible by streamlining the pull of openly available online data into a machine-readable dataset. AI/ML will also support the MedINT Canvas visuals by finding relevant terms in online articles, labeling articles accordingly, and ranking them for relevance to topics of interest. These necessary steps cannot practically be performed by people. Allowing analysts to wholistically visualize health security data will enable identification of potential threats that should be further investigated. 

The main AI system output will be the translation of online information into a structured, machine-readable dataset. This includes language translation for non-English articles, labeling articles by health threat types, and ranking articles by relevance to health threats. This dataset will then be used to create the wholistic visuals in Canvas and Dashboard. 

MedINT will use only open-source and aggregated level data without personal identifiers.  It will not use individual medical data. 

Use Case Topic Areas: Health & Medical 

Deployment Status: Pre-deployment (Initiation) 

Safety- and/or rights-impacting? No, use case is too new to fully assess impacts; will be reassessed before end of initiation stage. 

Face Recognition/Face Capture (FR/FC)? No 

Use Case Name: One Health Threat Detection and Risk Assessment Platform (OH-TREADS)/Planner 

Use Case ID: DHS-2421 

Use Case Summary: The goal for this proof-of-concept project is to create wholistic visuals for health surveillance from relevant data and information sourced (open and non-public) from  International, Federal, State, Local, Tribal, and Territorial (IFSLTT) partners and stakeholders in interdependent sectors, including but not limited to the Food & Agriculture and Healthcare & Public Health sectors. By combining human, animal, plant, and environmental data, with critical infrastructure data, we aim to generate an integrated One Health data landscape that can aid decision making, early warning, and enhanced situational awareness. This would improve detection, prevention, and mitigation of emerging threats to health, food, and agriculture (such as Chemical, Biological, Radiological, and Nuclear (CBRN), climate, pests/vectors, invasive species, etc.) whether natural, intentional, or accidental, and promote health security, preparedness, and resilience across domains, disciplines, sectors, and stakeholders. OHTREADs will use a similar approach to Medical Intelligence (MedINT), employing a dashboard to present a telescoping global-to-local map that will highlight health incidents, cases, or other epidemiological relevant data that can be linked via geospatial or temporal factors. Upon clicking the map, the user will be able to achieve state or county level data visualization, or other relevant user-filtered information for active incidents involving pathogens/agents of interest, including risk and predictive transmission values. AI/ML makes this project possible by streamlining the pull of accessible data into machine-readable datasets from veterinary syndromic, clinical, diagnostic, or other surveillance data sets. AI/ML will also create the OH-TREADs/Planner visuals and allow for analytical interpretation. 

The volume of data, and lack of clear data collection standards, requires AI/ML to help merge streams with different data ontologies while facilitating data interoperability with mis-paired data sets. AI/ML will also be used to generate risk and predictive scores, displaying relevant information and analyses that help analysts understand health security threats and broadly monitor the health security landscape. This comprehensive situational awareness will support data-driven decision making to prevent, mitigate, and respond to health threats. 

The main AI system outputs will be anomaly detection following the translation of information into a structured, machine-readable datasets. AI/ML is then used for risk identification and disease prediction, which are overlayed on a map and displayed with other wholistic visuals. These visuals and analytics can be combined with critical infrastructure locations, resources, or capabilities (federal, state, local, tribal, and territorial) to aid response and decision making. 

OH-THREADS will use a combination of open source and public health data sets as well as animal health records.  It will not use individual medical data. 

Use Case Topic Areas: Health & Medical 

Deployment Status: Pre-deployment (Initiation) 

Safety- and/or rights-impacting? No, use case is too new to fully assess impacts; will be reassessed before end of initiation stage. 

Face Recognition/Face Capture (FR/FC)? No