Google Opens Access to Med-Gemma 1.5 & Med-ASR AI Models for Healthcare

Key Takeaways:

• Google has released two open-access AI models for healthcare: Med-Gemma 1.5 for medical imaging and Med-ASR for speech recognition.
• The models aim to accelerate research in medical diagnosis and clinical documentation.
• Both are available now for developers but are not yet certified for clinical use.

Google has made two specialised artificial intelligence models openly available to researchers and developers, targeting critical areas in healthcare. The release includes Med-Gemma 1.5 for analysing medical images and Med-ASR for transcribing medical conversations.

In an official announcement, Google stated: “Today, we’re releasing two new models to the research community: Med-Gemma 1.5, a state-of-the-art open model for medical imaging, and Med-ASR, a family of open models for medical speech recognition.” The move aligns with the company’s commitment to open science and responsible AI development.

Med-Gemma 1.5: The Medical Imaging Specialist

Built on Google’s Gemma 2 language models, Med-Gemma 1.5 is fine-tuned for radiology. It can analyse X-rays and CT scans, generate reports, and answer questions about the imagery. Google claims it outperforms other leading open models on medical imaging benchmarks.

The model comes in two variants: a 2-billion parameter version and a more powerful 9-billion parameter version. Both are released under an open license permitting commercial use, available on platforms like Hugging Face and Kaggle.

Med-ASR: Tackling Medical Speech Transcription

Transcribing doctor-patient dialogues or clinical notes is challenging due to complex terminology. Med-ASR is a family of models trained on millions of hours of medical audio to accurately understand medical terms, abbreviations, and clinical dialogue flow.

The initial release supports English and is designed to improve the accuracy of medical speech recognition, a common pain point in healthcare documentation .

Access and Future Impact

Google’s open release is designed to spur innovation in healthcare AI. By providing these tools, the company hopes developers will create applications that enhance diagnosis, streamline documentation, and improve patient outcomes.

While Med-Gemma 1.5 is on Hugging Face and Kaggle, Med-ASR is hosted on GitHub. Google has clarified that these are currently for research and development only and are not yet approved for direct clinical application.