University hospital ranked among the 5 most technological and smart hospitals in Europe.
Health
AI Agent for clinical information processing and structuring
In the healthcare field, hospitals and medical centers generate daily large volumes of clinical information of enormous value for care, research, and population analysis.
A large part of this information is found in free text, such as clinical notes, hospital reports, emergency records, or clinical courses. Abbreviations, different writing styles, and multiple ways of expressing the same reality make automatic processing and reuse difficult.
In this context, the hospital needed to advance in the structuring and normalization of clinical information, leveraging the capabilities of agentic AI without losing traceability, the original context of each data point, or control over especially sensitive healthcare information.
Challenge
The main challenge was to transform large volumes of unstructured clinical text into homogeneous and processable information, without losing the necessary context to interpret and validate each result.
The solution had to automatically identify diagnoses and procedures, recognize attributes such as temporality, certainty, negation, or subject, and link each result with the textual evidence supporting it. All of this had to be carried out within the hospital’s infrastructure, preventing clinical information from leaving its environment.
Solution
Foqum has developed an AI agent based on an agentic AI architecture capable of processing complete clinical episodes and transforming free text into structured, normalized, and traceable information.
The system coordinates different extraction, resolution, and coding stages to identify clinical entities and preserve, for each of them, the original phrase, its position within the note, and attributes such as temporality, certainty, negation, or subject. Subsequently, entities can be coded using standards such as ICD-10, SNOMED CT, or other vocabularies defined by the organization, facilitating interoperability and the subsequent exploitation of data.
The solution runs on-premise through AI models deployed directly within the hospital infrastructure, maintaining privacy and data control. The pipeline has reached a latency of 5.29 seconds per note in a production environment, demonstrating the feasibility of applying agentic AI to clinical processing efficiently and enabling new capabilities for research, population analysis, and interoperability.
