Radiology AI makes consistent diagnoses using 3D images from different health centres

Key Points

• The aim is to assess the performance of the Merlin AI model for diagnosing images from different hospitals.
• Train a 3D vision-language model called Merlin using abdominal computed-tomography scans, radiology reports, and electronic health records.
• Evaluate Merlin's performance compared to other vision-language models across three distinct hospital sites.
• Measure performance improvement in accurate diagnoses.
• Merlin outperforms other vision-language models by an average of 20% across different hospital sites.
• Demonstrates consistent diagnostic accuracy using 3D images from various health centres.