Med Gemini: A Step Forward

Med Gemini is revolutionizing the way we use ai to interact with medical data. In a study published by Google research in collaboration with Google DeepMind. This study showed us the enhanced capabilities of the multimodal models Med-Gemini-2D, Med-Gemini-3D and Med-Gemini Polygenic. Med Gemini was able to outperform GPT-4 when it came to a variety of tasks by about a margin of 44.5%

This study revealed that Med-Gemini-2D was able to perform better that expected and was as good or in some cases better the radiologist’s original reports. Med-Gemini-3D was able to read CT-scans and give reports and answer questions based on them. Out of these reports 53% of these were deemed clinically acceptable. Med-Gemini Polygenic is Googles model that uses genomic data. It is able to outperform standard approaches for disease and risk prediction and it is even able to generalize diseases for which it hasn’t been trained. Speaking of training these models were trained on a mixture of public and private datasets of about & million samples and 3.7 million medical images and cases.

While having conversations with Med Gemini it was able to display a wide range of knowledge based on images provided. While looking at real world applications it was able to correctly diagnose the patient with a provided picture. While not perfect it is able to make quick and accurate predictions based on the given data.

All this marks that we are a step closer to being able to depend on these tools to aid are doctors and speed up all medical diagnoses. While this being significant progress there is still a long way before we reach a seamless integration of AI in all healthcare scenarios. Soon misdiagnoses will be a thing of the past and will usher in a new age of medical advancement.

References

Advancing Multimodal Medical Capabilities of Gemini - Google DeepMind

An AI Club Article by Sidhant Nair