ItalyHealthcare and life sciences2022 – ongoingIEO-Monzino
Language model that turns medical reports into structured research data
IEO-Monzino structured medical reports 300 times faster than manual work with its own language model
Problem
The European Institute of Oncology and the Monzino Cardiology Center in Milan hold large volumes of free-text reports and exams. Researchers needed them as standardized, anonymized data. One physician had extracted biomarkers and diagnoses by hand from 500 pathology reports, and that work took six months.
Approach
The hospitals began using Google Cloud in August 2022 to build a Clinical Data Platform. The IT team built its own natural language processing model on Vertex AI, using the manual extraction work as a starting point. It collected structured information from 76,000 medical reports. The team had developed two further models for blood-cell pathology reports, one for standardized diagnoses and one for biomarkers, and said four AI models were in development.
Results
Takeaway
A manual extraction exercise on a small set of reports gave the team labelled examples for its model. The same method was planned for breast radiology reports.
Categories
- Industry
- Healthcare and life sciences
- Business function
- IT and R&D
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
Google Cloud's customer story describes how IEO-Monzino built a natural language processing model on Vertex AI to extract structured data from medical reports for research. It reports that 76,000 reports were processed and that processing ran 300 times faster than the earlier manual work. The page carries no date; the Internet Archive first saved it on May 30, 2024.
Full details are in the original sources.
