Google shares that Med-Gemini, an advanced AI model specialized in medicine, is more factually accurate, reliable, and produces nuanced results for complex clinical tasks than OpenAI’s GPT-4.
DeepMind, Google’s AI research lab and Google Research released a recent paper on their upcoming AI tool for use in the healthcare space.
Still in the research phase, Med-Gemini has been equipped with the latest technology that can even surpass popular industry standards, according to Google researchers.
Med-Gemini comes with large multimodal models (LMMs) and they are all designed for different purposes and applications.
Google’s Gemini models were by default well armed with advanced technologies. They could process information from text, images, video and audio. This makes Med-Gemini far more efficient as it’s fine-tuned with all these specialties.
Some of these include:
Web search capabilities and ability to self-train
Med-Gemini is capable of accessing web-based searching that will help in enhancing advanced clinical reasoning. Med-Gemini has been tested on 14 medical benchmarks and has established a new state-of-art (SOTA) performance on 10. Its performance has surpassed the GPT model family on every aspect where a comparison can be made.
On the MedQA (USMLE) benchmark, Med-Gemini achieved 91.1% accuracy using its uncertainty-guided search strategy, outperforming Google’s previous medical LLM, Med-PaLM 2, by 4.5%, according to New Atlas.
Easily analyzing lengthy electronic health records
Electronic health records (EHRs) are typically very long and contain several similarities in text. Med-Gemini can help with handpicking relevant information from lengthy documents. Researchers wanted to expound on this capability of Med-Gemini and ran a so-called ‘needle in the haystack’ task.
They used a huge publicly available database, the Medical Information Mart for Intensive Care or MMIC-III, which consisted of the deidentified health data of patients who were admitted to intensive care. The main purpose behind the task was to pick out a relevant mention of a rare and subtle medical condition, symptom, or procedure over a huge collection of data in the EHR.
Med-Gemini performed quite well in the test. It had to carefully hand-pick all mentions of the specified medical problem from the health records. Besides this, it also had to evaluate how these mentions were relevant, categorize them and reveal whether a particular patient had a history of that problem, and also showcase its reasoning.
“Perhaps the most notable aspect of Med-Gemini is the long-context processing capabilities because they open up new performance frontiers and novel, previously infeasible application possibilities for medical AI systems,” said the researchers.
The future
Amid all the positivity centered around Med-Gemini, what does the future look like? In the words of the researchers, there’s a lot to be done. However, Med-Gemini has definitely showcased a promising future for itself.
Besides this, Google plans to abide by fairness and privacy throughout the development process. To quote the researchers, “Privacy considerations in particular need to be rooted in existing healthcare policies and regulations governing and safeguarding patient information.”
“Fairness is another area that may require attention, as there is a risk that AI systems in healthcare may unintentionally reflect or amplify historical biases and inequities, potentially leading to disparate model performance and harmful outcomes for marginalized groups,” they added.
