Google has released a significant update to its suite of open-source medical AI models, highlighting the healthcare industry's rapid adoption of artificial intelligence. The announcement of MedGemma 1.5 and a new medical speech-to-text model, MedASR, follows a broader trend of major AI companies deepening their investment in health technology.

Enhanced Medical Imaging Interpretation

The updated MedGemma 1.5 model introduces expanded capabilities for interpreting complex medical imagery, building on its existing functionality for 2D images like X-rays. Key new features for developers include:

1. Support for 3D and High-Dimensional Scans: The model can now process volumetric data from CT and MRI scans, as well as whole-slide histopathology images.
2. Longitudinal Analysis: Improved ability to review time-series data, such as sequences of chest X-rays, to track changes over time.
3. Anatomical Localization: Enhanced precision in identifying and localizing specific anatomical features within images.

Google reports internal benchmark improvements, including a 14% increase in accuracy for classifying disease-related MRI findings compared to the previous model version.

New Medical Speech-to-Text Model

Alongside the image model update, Google introduced MedASR, an automated speech recognition model fine-tuned for medical dictation and vocabulary. The company states MedASR can be used to transcribe clinician notes or serve as a voice interface to prompt the MedGemma model, reducing word error rates significantly compared to general-purpose speech models on medical audio.

Developer Focus

This release follows a pattern of heightened activity in AI health infrastructure, coming shortly after OpenAI's move to integrate the health data startup Torch. These parallel developments by leading labs highlight a concerted industry effort to address long-standing challenges like clinical data fragmentation and workflow support, which could lead to rapid improvements in how doctors analyze information and manage patient care.

Google's models are released under its Health AI Developer Foundations (HAI-DEF) program as open-source starting points for developers and researchers. To encourage innovation, the company is launching the "MedGemma Impact Challenge," a Kaggle hackathon with $100,000 in prizes.

Access and Context

MedGemma 1.5 and MedASR remain free for research and commercial use and are available on Hugging Face and Google's Vertex AI platform. The release underscores the ongoing push by leading AI labs to provide the underlying infrastructure for a new generation of AI-assisted healthcare tools, a sector seeing intensified activity and investment.

Links:

Hugging face:

Vertex:

OpenAI: