MedGemma 1.5 is expanding support for high-dimensional medical imaging, adding 3D volume representations of CT imaging and MRI, along with whole-slide histopathology imaging. The update builds on MedGemma 1, which supported 2D medical images such as chest X-rays, dermatology images, fundus images and histopathology patches.
The model is designed so developers can provide multiple slices for CT or MRI, or multiple patches for histopathology, together with a prompt describing the task. Google says this makes MedGemma 1.5 a multimodal model that can handle both high-dimensional medical data and general 2D data and text.
On internal benchmarks, MedGemma 1.5 improved baseline absolute accuracy by 3% over MedGemma 1, from 58% to 61%, on classification of disease-related CT findings. It also improved by 14%, from 51% to 65%, on classification of disease-related MRI findings, averaged over findings.
On an internal diverse benchmark of histopathology slides and associated findings, the fidelity of MedGemma 1.5’s predictions, based on ROUGE-L score on cases with exactly one histopathology slide, improved from 0.02 to 0.49. Google says this matched the 0.498 score achieved by the task-specific PolyPath model.
Google described the new capability as the natural evolution of CT foundation, its earlier API-based tool for generation of CT embeddings. It also said that, to its knowledge, MedGemma 1.5 is the first public release of an open multimodal large language model that can interpret high-dimensional medical data while retaining support for general 2D data and text.
The company said the capabilities are still in early stages and remain imperfect, and that developers will achieve improved results by fine-tuning MedGemma models on their own data. Tutorial notebooks have been released for CT and histopathology use cases on Hugging Face and Model Garden.
Source: research.google.
Companies can share verified announcements through Newz9’s international press release submission page.
