MedGemma Notebooks

Quick start with Hugging Face - Example of generating responses from medical text and images by running the model locally from Hugging Face.
Quick start with Model Garden - Example of serving the model on Vertex AI and using Vertex AI APIs to generate responses from medical text and images in online or batch workflows.
Fine-tune with Hugging Face - Example of fine-tuning the model with Low-Rank Adaptation Supervised Fine-tuning (SFT) and LoRA using Hugging Face libraries.
Quick start with Digital Imaging and Communications in Medicine (DICOM) - Example of using DICOM-aware MedGemma deployments which can handle medical image retrieval and pre-processing on the server side.
Electronic Health Record (EHR) Navigator Agent - Demonstration of an AI agent using MedGemma to intelligently navigate patients' large and complex EHR data in Fast Healthcare Interoperability Resources (FHIR) format.
Reinforcement learning with Hugging Face - Example of fine-tuning the model using reinforcement learning, a tuning method of tuning that is particularly effective at learning complex tasks without compromising existing model abilities.

New with MedGemma 1.5

Chest X-ray anatomy localization - Illustrates how to perform a localization task with MedGemma 1.5 using Hugging Face libraries and running the model locally.
Chest X-ray longitudinal comparison - Walks through how to perform longitudinal comparison of chest X-ray (CXR) images with MedGemma 1.5 using Hugging Face libraries and running the model locally.
Analyze Computed Tomography (CT) Imaging - Hugging Face - Explores how 3D representation of computed tomography (CT) imaging can be used to prompt a locally running MedGemma 1.5 using the Hugging Face libraries.
Analyze Computed Tomography (CT) Imaging - Model Garden - Showcases how 3D representation of computed tomography (CT) imaging can be used to prompt MedGemma 1.5 running on Vertex AI.
Analyze Whole Slide Digital Pathology Imaging - Hugging Face - Demonstrates how to use digital pathology whole slide imaging to prompt a locally running MedGemma 1.5 using the Hugging Face libraries.
Analyze Whole Slide Digital Pathology Imaging - Model Garden - Presents how to use digital pathology whole slide imaging to prompt MedGemma 1.5 running on VertexAI.

Name		Name	Last commit message	Last commit date
parent directory ..
README.md		README.md
cxr_anatomy_localization_with_hugging_face.ipynb		cxr_anatomy_localization_with_hugging_face.ipynb
cxr_longitudinal_comparison_with_hugging_face.ipynb		cxr_longitudinal_comparison_with_hugging_face.ipynb
ehr_navigator_agent.ipynb		ehr_navigator_agent.ipynb
evaluate_on_medqa.ipynb		evaluate_on_medqa.ipynb
fine_tune_with_hugging_face.ipynb		fine_tune_with_hugging_face.ipynb
high_dimensional_ct_hugging_face.ipynb		high_dimensional_ct_hugging_face.ipynb
high_dimensional_ct_model_garden.ipynb		high_dimensional_ct_model_garden.ipynb
high_dimensional_pathology_hugging_face.ipynb		high_dimensional_pathology_hugging_face.ipynb
high_dimensional_pathology_model_garden.ipynb		high_dimensional_pathology_model_garden.ipynb
quick_start_with_dicom.ipynb		quick_start_with_dicom.ipynb
quick_start_with_hugging_face.ipynb		quick_start_with_hugging_face.ipynb
quick_start_with_model_garden.ipynb		quick_start_with_model_garden.ipynb
reinforcement_learning_with_hugging_face.ipynb		reinforcement_learning_with_hugging_face.ipynb