An AI assistant for oncology teams that processes documentation, prepares materials to support decision-making, and facilitates the organisation and minutes of MDT meetings.
Preparing for a multidisciplinary oncology team (MDT) meeting can be very time-consuming. The doctor must go through extensive medical records, reconstruct the diagnostic and treatment process, and prepare a concise yet clinically relevant summary. This can take up to 20 minutes per patient, whilst 20–40 patients are typically discussed at a single MDT meeting. This often leaves only a few minutes for the actual clinical decision-making.
We have therefore developed a prototype AI assistant to help with the preparation, organisation and documentation of MDTs.
The application enables the management of a patient database, the uploading of medical records and their automatic processing into a structured format. This includes the anonymisation of records, quality control and the identification of missing or unclear data. A clear clinical summary and an interactive timeline of diagnosis and treatment are generated for each patient. Individual statements can be traced back to the original documents, enabling doctors to quickly verify the source of the information.
However, we do not wish to limit ourselves to merely summarising the records.
The concept also includes an analysis of care provided to date in relation to professional guidelines, the identification of potential deviations, and the proposal of further treatment options with references to relevant guidelines. Another module utilises prognostic data published in clinical trials and enables this to be placed within the context of the treatment options under consideration.
The application also includes an MDT calendar, patient scheduling and the management of participants in individual meetings. We also plan to support automatic meeting minutes using AI, including the identification of individual speakers, the structuring of conclusions and the creation of tasks for the oncology coordinator.
Our aim is not to replace clinical decision-making, but to provide doctors with well-prepared, traceable and relevant information at the very moment they need it most.
We have developed and tested the prototype using data from the field of neuro-oncology; however, its potential extends well beyond this specialism. The proposed solution can be applied across a range of oncological diagnoses and, in the future, in other medical disciplines where multidisciplinary decision-making forms part of clinical practice. The next step involves testing in a real-world clinical setting, clinical validation of individual AI functions, evaluation of actual time savings, and assessment of regulatory requirements, including potential certification under the European Medical Devices Regulation (MDR). Our long-term goal is the safe and reliable implementation of the system into routine clinical practice.
- Had before
- Před zahájením projektu jsme neměli žádný existující kód, datovou sadu ani prototyp. Začínali jsme od nuly. Vycházeli jsme pouze z vlastní zkušenosti s každodenní klinickou praxí a znalosti problémů, se kterými se lékaři při přípravě multidisciplinárních týmů setkávají.
- Use of AI
- Rozne:
- Retrospective guidelines deviations - LLM (GPT 6 Astra; Azure) is given patient's conditions and documents, guidelines to follow, and is asked to look for differences.
- Suggested next steps - LLM (GPT 6 Astra; Azure) is given patient's conditions and documents, guidelines to follow, and asked for next steps.
- Prognosis - LLM (GPT 6 Astra; Azure) is given patient's conditions and documents, guidelines to follow, and searches research papers for
- Document text extraction - Local OCR as first pass and detection and censoring of sensitive data. Second pass sent to vision language model.
- Speech-to-text - AI meeting recorder and auto-summary - Live speech transcribed with ElevenLabs (demo). After the meeting is done, the conversation is summarized into 1-2 actionable outputs with LLMs (GPT 6 Luna; Azure) - Impact
- Naším cílem není jen ušetřit lékařům čas, ale zlepšit kvalitu rozhodování o onkologické léčbě. Při 20-40 pacientech na jednom MDT může i úspora 5-10 minut na případ znamenat 1,7-6,7 hodiny přípravy na jediné zasedání (zatím odhad).
Důležitější je ale klinický potenciál: úplnější podklady, dohledatelné zdroje a upozornění na možné odchylky od doporučených postupů mohou přispět k informovanějším a konzistentnějším rozhodnutím založeným na důkazech.
V dlouhodobém horizontu chceme ověřit, zda nástroj pomáhá snižovat počet neodůvodněných odchylek od doporučených postupů, zpoždění v péči a zlepšovat výsledky léčby, případně i přežití.
Pokud prokážeme bezpečnost a přínos, řešení lze rozšířit z neuroonkologie na další diagnózy a postupně do nemocnic napříč ČR.
Naší ambicí je, aby kvalita přípravy MDT nezávisela na tom, v jaké nemocnici se pacient léčí. - Who benefits
- Patients and their families, Doctors and nurses, Hospitals and care providers
- InterSystems technologies
- Yes — Importuje aj exportujeme pacientske data informácie.
Machine translation from Czech
- Team
- Codebridge — Soňa Hudecová, Juraj Oravec, Imrich Berta
- Challenge
- AI assistant for the multidisciplinary team · Ostrava