# MedGemma 1.5 expanded a 4B model toward volumes, slides and longitudinal imaging inputs Google released MedGemma 1.5 4B on January 13, 2026, adding stated support for CT and MRI volumes, histopathology patches, chest X-ray comparison and lab-report extraction. Canonical: https://rangoon.ai/insights/archive-google-medgemma-1-5-imaging-january-2026/ Author: Rangoon Editorial (https://rangoon.ai/insights/editorial/) Published: 2026-10-24 Updated: 2026-10-24 Event date: 2026-01-13 Tags: [Google](https://rangoon.ai/insights/tags/google/), [MedGemma 1.5](https://rangoon.ai/insights/tags/medgemma-1-5/), [medical imaging](https://rangoon.ai/insights/tags/medical-imaging/), [multimodal AI](https://rangoon.ai/insights/tags/multimodal-ai/) Source note: Historical reporting: Google Research released MedGemma 1.5 4B on January 13, 2026 and described high-dimensional imaging, pathology, longitudinal X-ray, localization and lab-report capabilities. The accompanying Google model card details inputs, preprocessing and intended-use limits. Reported benchmark deltas are Google baselines, not proof of clinical performance, safety, clearance or a recommendation for direct patient care. Image: High-dimensional imaging input assembly. Original Rangoon explanatory diagram: Source boundary: dated primary release reporting; original editorial analysis. No vendor interface, photograph, measurement result, or endorsement implied. ## Key takeaways - MedGemma 1.5 extended a 4B multimodal model from single 2D images toward volume and multi-patch inputs. - CT, MRI and pathology inputs require preprocessing choices that are part of the system contract. - Vendor-reported benchmark gains did not establish performance for another site, modality mix or workflow. ## The January release named a 4B update and new input classes [source 1](https://research.google/blog/next-generation-medical-image-interpretation-with-medgemma-15-and-medical-speech-to-text-with-medasr/) On January 13, 2026, Google Research announced MedGemma 1.5 4B. It described support for high-dimensional CT and MRI representations, whole-slide histopathology through multiple patches, longitudinal chest X-ray review, anatomical localization and structured extraction from laboratory reports. The event was a model update rather than a new imaging service or a clinical product announcement. Google described the model as a compute-efficient development starting point and kept the earlier 27B model for more complex text-oriented work. That division makes model selection part of a system design choice rather than a simple version-number upgrade. ## Volumes and slides required an input-assembly layer before language generation [source 1](https://research.google/blog/next-generation-medical-image-interpretation-with-medgemma-15-and-medical-speech-to-text-with-medasr/) [source 2](https://developers.google.com/health-ai-developer-foundations/medgemma/model-card) A CT or MRI study is not one ordinary two-dimensional image, and a pathology slide may be represented through many selected patches. Google’s announcement said developers could supply multiple slices or patches with a task prompt, while the model card notes that CT, MRI and whole-slide inputs require preprocessing. This means an application has to decide how to select, order, transform and retain those inputs before model inference. A result can depend on that assembly policy as well as on the model weights. The release therefore made a fuller pipeline visible: source data, preprocessing, multimodal prompt, generated output and task-specific evaluation. ## The reported gains were baseline comparisons under Google’s evaluations [source 1](https://research.google/blog/next-generation-medical-image-interpretation-with-medgemma-15-and-medical-speech-to-text-with-medasr/) [source 2](https://developers.google.com/health-ai-developer-foundations/medgemma/model-card) Google reported internal benchmark differences against MedGemma 1 4B for CT findings, MRI findings, histopathology, chest X-ray tasks and laboratory-report extraction. Those numbers identify what Google measured on its datasets and definitions; they do not guarantee a gain for a different scanner, acquisition protocol, patient population, prompt, annotation practice or data-quality regime. The model card similarly distinguishes baseline evaluation from the fine-tuning developers are expected to perform. A technical review should preserve the exact benchmark and version with a claim, then test the selected workflow rather than transferring a headline percentage into an unrelated environment. ## A richer modality interface did not turn output into a clinical decision [source 1](https://research.google/blog/next-generation-medical-image-interpretation-with-medgemma-15-and-medical-speech-to-text-with-medasr/) [source 2](https://developers.google.com/health-ai-developer-foundations/medgemma/model-card) MedGemma 1.5 could accept more complex inputs, but each capability remained an invitation to adapt and validate a downstream application. The documentation says outputs are not intended to directly inform diagnosis, patient-management decisions or treatment recommendations, and it notes that inaccurate outputs remain possible. That limitation is material to the release story because more input modalities can create a stronger impression of completeness without resolving missing context, incorrect preprocessing or model error. The historical conclusion is limited: Google released a 4B update with stated imaging and document capabilities. Whether any specific workflow is accurate, safe or suitable depends on evidence outside the announcement and model card. Comparing versions also requires holding the input-assembly policy fixed; otherwise an observed difference can be caused upstream of the model. ## Sources 1. [Google Research: MedGemma 1.5 and MedASR announcement](https://research.google/blog/next-generation-medical-image-interpretation-with-medgemma-15-and-medical-speech-to-text-with-medasr/) 2. [Google for Developers: MedGemma 1.5 model card](https://developers.google.com/health-ai-developer-foundations/medgemma/model-card) ## Related reading - [Google introduced MedGemma at I/O as an open starting point for medical text and image work](https://rangoon.ai/insights/archive-google-medgemma-health-models-may-2025/) - [Gemma 3n framed on-device multimodality as an artifact-provenance problem](https://rangoon.ai/insights/archive-google-gemma-3n-may-2025/) - [Liquid AI released LFM2-VL as compact vision-language models for edge deployment](https://rangoon.ai/insights/archive-liquid-lfm2-vl-edge-august-2025/) ## More information - [Documentation index](https://rangoon.ai/llms.txt) - [AI access and policies](https://rangoon.ai/ai/) - [Source repository](https://github.com/hypler-dev/rangoon)