The solution uses artificial intelligence to automatically extract and process data for the Czech National Cancer Registry (NOR). The team secured second place in the national finals and is now working on further developing the project.
Every newly diagnosed cancer patient must be registered in the Czech National Cancer Registry. Healthcare professionals currently have to review extensive medical documentation and manually enter the required information into a form. Processing a single case takes approximately 20 to 30 minutes. With around 65,000 new cancer cases diagnosed annually in the Czech Republic, this amounts to tens of thousands of hours of administrative work.
Boris Pustějovský, Dominik Vilímek and Jindřich Brož, who make up team ICU, set out to simplify this process. Their application, ONTRA, automatically analyses medical reports, identifies information about diagnoses, disease stages and treatments, and pre-fills the NOR registration form. Each extracted piece of information can be traced back to its source in the original documentation. If information is missing or reports contain conflicting data, the application flags the issue and leaves the final decision to a healthcare professional.
“We wanted to create a tool that would genuinely make healthcare professionals' work easier while keeping them in control of the results. Instead of spending time manually entering data, they can simply verify the pre-filled information. A process that currently takes up to half an hour could be reduced to approximately two minutes. Healthcare professionals could then devote the time saved directly to patient care,” explains Dominik Vilímek.
An Application Developed in Just 24 Hours
During the hackathon, the team developed a functional application capable of processing medical documentation, including photographs of medical reports. One of the biggest challenges was ensuring that the application could run directly within a hospital, without relying on external cloud services, so that sensitive patient data would not have to leave the hospital's infrastructure. The application's outputs also comply with the NOR format and the international HL7 FHIR standard, which enables the exchange of healthcare information between systems.
The application achieved an accuracy of over 93% on an available test dataset of real patient records. The team also reported no instances in which the system incorrectly marked uncertain information as verified. To support further testing, the developers created a synthetic data generator, producing records for nearly 2,000 fictional patients and approximately 8,800 medical reports.
“The hardest part was developing a functional solution in a single day while also addressing the security and operational requirements of a hospital environment. We opted for the technically more demanding approach of running the application locally and developed our own testing method at the same time. We didn't want to present just a demonstration of what artificial intelligence can do, but an application with genuine potential for practical use,” says Boris Pustějovský.
Bringing Together Three Technical Disciplines
The combination of expertise in computer science, cybernetics and biomedical engineering — all fields taught at the Faculty of Electrical Engineering and Computer Science — played a significant role in the project's development. While computer science and cybernetics contributed to system design, data processing and algorithms, biomedical education and experience in radiology helped the team understand medical documentation and the clinical significance of individual data points.
“When developing healthcare applications, it is not enough for the software to simply work from a technical perspective. You also need to understand how healthcare professionals use information and what consequences may arise if that information is interpreted incorrectly. Bringing together our different fields of expertise enabled us to develop a solution that addresses both technical and clinical requirements,” adds Jindřich Brož.
From Competition Prototype to Hospital Use
The team sees its second-place finish in the national Rakathon finals, accompanied by a prize of CZK 50,000, primarily as motivation to continue developing the project. Their next step is to validate the application using a larger dataset of real patient records and across multiple healthcare facilities.
The developers are building on their collaboration with University Hospital Ostrava and are also in discussions with other healthcare providers. In the future, they aim to expand the system to cover additional diagnoses, automate the processing of radiology reports and achieve deeper integration with hospital information systems.
“For us, Rakathon is not the end, but the beginning. The project will only truly fulfil its purpose when it helps the people who currently have to manually transcribe information from medical records. That's what we're working towards now,” the members of team ICU conclude.