A study led by researchers from Mohammed bin Zayed University has introduced an artificial intelligence model named BiMediX2. This system is capable of understanding and interpreting information in medical images, supporting bilingual analysis in both English and Arabic, with particularly outstanding performance in processing Arabic content.
BiMediX2 is the first artificial intelligence system that can analyze and describe medical images in both English and Arabic. The system can handle various types of medical images, including X-rays, MRI scans, and microscopic images, providing detailed descriptions and answering questions related to the image content. This model not only improves the accuracy of translations in understanding medical images but also significantly enhances the processing capabilities for Arabic.
According to the technical report, BiMediX2 has shown excellent performance in comparative tests against existing technologies, improving the translation accuracy for English content by 9% and for Arabic content by 20%. This achievement is backed by the team's extensive training on 1.6 million medical texts and image data, ensuring accurate conversion between the two languages.
BiMediX2's exceptional performance is attributed to innovations in its underlying architecture. The system is based on the Llama3.1 architecture and GPT-4o, specifically optimized for the medical field. By integrating with Vision Encoder and Meta Llama3.1, BiMediX2 enables seamless bilingual medical image analysis. In tests, it has shown a greater advantage over GPT-4o in identifying incorrect medical information.
Despite BiMediX2's promising performance, researchers emphasize that the system is currently limited to research purposes and has not yet entered clinical applications. Like all artificial intelligence systems, BiMediX2 may also make mistakes or generate inaccurate information. Therefore, the research team has released the model on Hugging Face and launched a bilingual benchmark named BiMed-MBench to test the performance of similar systems.
BiMediX2 represents a significant innovation in the field of medical image analysis, particularly achieving notable progress in handling bilingual medical images and translations. Although it is not yet applied in clinical settings, the research findings of this system promise more efficient and accurate applications of artificial intelligence in the healthcare industry.
