A recent study demonstrates that free, open-source AI tools can perform as well as, or even better than, expensive commercial systems in medical scan reporting. These tools offer a cost-effective alternative to options like ChatGPT, without compromising patient privacy.
The study focused on thyroid nodules, utilising the ACR TI-RADS scoring system to evaluate cancer risk. Researchers trained AI models using 3,000 synthetic radiology reports to avoid using real patient data. One open-source model, Yi-34B, matched the performance of GPT-4 when provided with learning examples.
AI tools are increasingly used to structure doctors' notes from medical scans, improving efficiency in tracking patient outcomes and spotting trends. Open-source AI models can be used within hospital systems, eliminating the need to send data to external servers and ensuring patient data protection.