top of page
Screenshot 2024-07-18 at 15.12.43 1

Blog post

University of Thessaly: AI Toolkit for Emergency Response: Reinforcing the Line Between Machine Intelligence and Emergencies Management

October 8th, 2026

by UTH

When emergencies unfold, information rarely arrives neatly

ESCORT_BlogImage.png

In an emergency, the challenge is not simply a lack of information. Quite often, it is the opposite: information arrives continuously, from different people, systems and data sources and must be understood quickly enough to support actions. Somewhere else, unusual patterns may already be appearing across population behaviour, healthcare activity or mobility data.

​

In both cases, valuable signals exist, but they are fragmented, time-sensitive and difficult to process manually.

This is where AI can support emergency response: not by replacing the professionals responsible for critical decisions, but by helping them to transform complex information into something that can be interpreted and acted upon.

From spoken words to structured clinical information

During pre-hospital care, clinical information is often generated naturally through conversation: symptoms are described, medications are administered, vital signs are reported and decisions are made.

 

The Speech-to-Data system that UTH developed within ESCORT is designed to capture these data without forcing emergency professionals to stop what they are doing and manually structure every piece of information.A paramedic can speak naturally while the system transcribes the conversation in real time.

 

From that transcript, relevant clinical information is extracted and validated before being transformed into a standardised HL7 FHIR representation that can be used by downstream healthcare systems. The solution also maintains links between the structured information and the original transcript, helping preserve traceability.

 

This is particularly important in emergency environments, where speed matters but incorrect information can be costly. The system, therefore, does more than transcription: extracted information is checked against the original conversation before being accepted as clinical data.

 

In simple terms, AI helps convert what was said during an incident into information that a digital healthcare systems can understand.

Picture 1.png

From scattered population signals to an early warning

A novel pathogen with pandemic potential may begin spreading before traditional surveillance systems have enough confirmed cases to clearly identify the threat. Early warning signs may instead appear indirectly across several sources.

 

UTH developed an AI surveillance approach looks for such signals by combining specific data. Different machine-learning methods independently search for deviations from expected behaviour, and their outputs are combined into an overall risk assessment mechanism.

 

Rather than asking the AI to diagnose a new pandemic, the goal is more focused: identify when several independent indicators begin behaving unusually enough to deserve attention.

 

This distinction is important. An anomaly is not automatically an emergency. It is a signal that can help public-health experts decide where closer investigation may be needed.

UTH ELLIE BLOG Post 2.jpg

Human judgement remains at the centre of decision making

AI can process emergency-related information at a speed and scale that would be difficult to achieve manually. But the interpretation of what these signals mean and the decision about what happens next remains a human responsibility.

 

The value of AI in emergency response lies not in replacing judgement, but in helping ensure that the right information reaches the right people at the right time.

bottom of page