Google's MedGemma 1.5: Open-Source AI for 3D Medical Imaging | Quick Digest

Google's MedGemma 1.5: Open-Source AI for 3D Medical Imaging | Quick Digest
Google has launched MedGemma 1.5, an updated open-source medical AI model capable of analyzing 3D CT and MRI scans. This advancement, alongside MedASR for medical speech-to-text, expands multimodal capabilities for global healthcare developers, offering improved diagnostic support.

Google releases MedGemma 1.5, an enhanced open-source medical AI model.

New version supports advanced 3D CT, MRI, and histopathology image analysis.

MedGemma 1.5 improves diagnostic accuracy and multimodal medical reasoning.

Accompanied by MedASR, a specialized medical speech-to-text transcription tool.

Models are free for global research and commercial development in healthcare.

Clinical deployment for diagnosis requires regulatory approval and validation.

Google has significantly advanced its commitment to open-source medical artificial intelligence with the release of MedGemma 1.5, an updated multimodal large language model. This new iteration greatly expands its capabilities to include the analysis of high-dimensional medical imaging, specifically three-dimensional Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) volumes, as well as whole-slide histopathology. Previously limited to 2D images and text, MedGemma 1.5 now processes entire scan volumes, enabling developers to build more comprehensive healthcare applications. The official launch, announced on January 13, 2026, marks a crucial step in making advanced AI tools accessible to the global healthcare community. MedGemma 1.5 improves diagnostic accuracy, with internal benchmarks showing a 14 percentage point increase in MRI classification accuracy to 65 percent and a 3 percentage point rise for CT classification to 61 percent. It also demonstrates enhanced performance in text-based tasks like medical reasoning and extracting information from electronic health records. Crucially, MedGemma 1.5 is released under Google's Health AI Developer Foundations (HAI-DEF) program, making it an open model, free for research and commercial use. Developers can download it from Hugging Face or adapt it through Google Cloud's Vertex AI. The release also introduces MedASR, a new open automated speech recognition model specifically fine-tuned for medical dictation, which has shown significantly fewer errors than general-purpose models like OpenAI's Whisper large-v3. While offering broad access, Google emphasizes that MedGemma 1.5 is intended as a starting point for developers and requires further fine-tuning and validation for specific clinical use cases. Its use for direct patient diagnosis or treatment is subject to regulatory approval as a medical device. The model's accessibility is expected to lower innovation barriers for medical institutions and startups worldwide, with mentions of adoption by entities like Tap Health in India for context-sensitive clinical tasks, highlighting its global relevance. This dual-model release underscores Google's strategy to foster an open and empowered ecosystem in medical AI, aligning with the accelerated adoption of AI in the healthcare industry.
Read the full story on Quick Digest