Buchumschlag

MedGemma Technical Report

Fuente: arXiv

Gespeichert in:

Bibliographische Detailangaben
Hauptverfasser:	Sellergren, Andrew, Kazemzadeh, Sahar, Jaroensri, Tiam, Kiraly, Atilla, Traverse, Madeleine, Kohlberger, Timo, Xu, Shawn, Jamil, Fayaz, Hughes, Cían, Lau, Charles, Chen, Justin, Mahvar, Fereshteh, Yatziv, Liron, Chen, Tiffany, Sterling, Bram, Baby, Stefanie Anna, Baby, Susanna Maria, Lai, Jeremy, Schmidgall, Samuel, Yang, Lu, Chen, Kejia, Bjornsson, Per, Reddy, Shashir, Brush, Ryan, Philbrick, Kenneth, Asiedu, Mercy, Mezerreg, Ines, Hu, Howard, Yang, Howard, Tiwari, Richa, Jansen, Sunny, Singh, Preeti, Liu, Yun, Azizi, Shekoofeh, Kamath, Aishwarya, Ferret, Johan, Pathak, Shreya, Vieillard, Nino, Merhej, Ramona, Perrin, Sarah, Matejovicova, Tatiana, Ramé, Alexandre, Riviere, Morgane, Rouillard, Louis, Mesnard, Thomas, Cideron, Geoffrey, Grill, Jean-bastien, Ramos, Sabela, Yvinec, Edouard, Casbon, Michelle, Buchatskaya, Elena, Alayrac, Jean-Baptiste, Lepikhin, Dmitry, Feinberg, Vlad, Borgeaud, Sebastian, Andreev, Alek, Hardin, Cassidy, Dadashi, Robert, Hussenot, Léonard, Joulin, Armand, Bachem, Olivier, Matias, Yossi, Chou, Katherine, Hassidim, Avinatan, Goel, Kavi, Farabet, Clement, Barral, Joelle, Warkentin, Tris, Shlens, Jonathon, Fleet, David, Cotruta, Victor, Sanseviero, Omar, Martins, Gus, Kirk, Phoebe, Rao, Anand, Shetty, Shravya, Steiner, David F., Kirmizibayrak, Can, Pilgrim, Rory, Golden, Daniel, Yang, Lin
Format:	Preprint
Veröffentlicht:	2025
Schlagworte:	Artificial Intelligence Computation and Language Computer Vision and Pattern Recognition
Online-Zugang:	Acceder al recurso
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

_version_	1866913008601530368
author	Sellergren, Andrew Kazemzadeh, Sahar Jaroensri, Tiam Kiraly, Atilla Traverse, Madeleine Kohlberger, Timo Xu, Shawn Jamil, Fayaz Hughes, Cían Lau, Charles Chen, Justin Mahvar, Fereshteh Yatziv, Liron Chen, Tiffany Sterling, Bram Baby, Stefanie Anna Baby, Susanna Maria Lai, Jeremy Schmidgall, Samuel Yang, Lu Chen, Kejia Bjornsson, Per Reddy, Shashir Brush, Ryan Philbrick, Kenneth Asiedu, Mercy Mezerreg, Ines Hu, Howard Yang, Howard Tiwari, Richa Jansen, Sunny Singh, Preeti Liu, Yun Azizi, Shekoofeh Kamath, Aishwarya Ferret, Johan Pathak, Shreya Vieillard, Nino Merhej, Ramona Perrin, Sarah Matejovicova, Tatiana Ramé, Alexandre Riviere, Morgane Rouillard, Louis Mesnard, Thomas Cideron, Geoffrey Grill, Jean-bastien Ramos, Sabela Yvinec, Edouard Casbon, Michelle Buchatskaya, Elena Alayrac, Jean-Baptiste Lepikhin, Dmitry Feinberg, Vlad Borgeaud, Sebastian Andreev, Alek Hardin, Cassidy Dadashi, Robert Hussenot, Léonard Joulin, Armand Bachem, Olivier Matias, Yossi Chou, Katherine Hassidim, Avinatan Goel, Kavi Farabet, Clement Barral, Joelle Warkentin, Tris Shlens, Jonathon Fleet, David Cotruta, Victor Sanseviero, Omar Martins, Gus Kirk, Phoebe Rao, Anand Shetty, Shravya Steiner, David F. Kirmizibayrak, Can Pilgrim, Rory Golden, Daniel Yang, Lin
author_facet	Sellergren, Andrew Kazemzadeh, Sahar Jaroensri, Tiam Kiraly, Atilla Traverse, Madeleine Kohlberger, Timo Xu, Shawn Jamil, Fayaz Hughes, Cían Lau, Charles Chen, Justin Mahvar, Fereshteh Yatziv, Liron Chen, Tiffany Sterling, Bram Baby, Stefanie Anna Baby, Susanna Maria Lai, Jeremy Schmidgall, Samuel Yang, Lu Chen, Kejia Bjornsson, Per Reddy, Shashir Brush, Ryan Philbrick, Kenneth Asiedu, Mercy Mezerreg, Ines Hu, Howard Yang, Howard Tiwari, Richa Jansen, Sunny Singh, Preeti Liu, Yun Azizi, Shekoofeh Kamath, Aishwarya Ferret, Johan Pathak, Shreya Vieillard, Nino Merhej, Ramona Perrin, Sarah Matejovicova, Tatiana Ramé, Alexandre Riviere, Morgane Rouillard, Louis Mesnard, Thomas Cideron, Geoffrey Grill, Jean-bastien Ramos, Sabela Yvinec, Edouard Casbon, Michelle Buchatskaya, Elena Alayrac, Jean-Baptiste Lepikhin, Dmitry Feinberg, Vlad Borgeaud, Sebastian Andreev, Alek Hardin, Cassidy Dadashi, Robert Hussenot, Léonard Joulin, Armand Bachem, Olivier Matias, Yossi Chou, Katherine Hassidim, Avinatan Goel, Kavi Farabet, Clement Barral, Joelle Warkentin, Tris Shlens, Jonathon Fleet, David Cotruta, Victor Sanseviero, Omar Martins, Gus Kirk, Phoebe Rao, Anand Shetty, Shravya Steiner, David F. Kirmizibayrak, Can Pilgrim, Rory Golden, Daniel Yang, Lin
contents	Artificial intelligence (AI) has significant potential in healthcare applications, but its training and deployment faces challenges due to healthcare's diverse data, complex tasks, and the need to preserve privacy. Foundation models that perform well on medical tasks and require less task-specific tuning data are critical to accelerate the development of healthcare AI applications. We introduce MedGemma, a collection of medical vision-language foundation models based on Gemma 3 4B and 27B. MedGemma demonstrates advanced medical understanding and reasoning on images and text, significantly exceeding the performance of similar-sized generative models and approaching the performance of task-specific models, while maintaining the general capabilities of the Gemma 3 base models. For out-of-distribution tasks, MedGemma achieves 2.6-10% improvement on medical multimodal question answering, 15.5-18.1% improvement on chest X-ray finding classification, and 10.8% improvement on agentic evaluations compared to the base models. Fine-tuning MedGemma further improves performance in subdomains, reducing errors in electronic health record information retrieval by 50% and reaching comparable performance to existing specialized state-of-the-art methods for pneumothorax classification and histopathology patch classification. We additionally introduce MedSigLIP, a medically-tuned vision encoder derived from SigLIP. MedSigLIP powers the visual understanding capabilities of MedGemma and as an encoder achieves comparable or better performance than specialized medical image encoders. Taken together, the MedGemma collection provides a strong foundation of medical image and text capabilities, with potential to significantly accelerate medical research and development of downstream applications. The MedGemma collection, including tutorials and model weights, can be found at https://goo.gle/medgemma.
format	Preprint
id	arxiv_https___arxiv_org_abs_2507_05201
institution	arXiv
publishDate	2025
record_format	arxiv
spellingShingle	MedGemma Technical Report Sellergren, Andrew Kazemzadeh, Sahar Jaroensri, Tiam Kiraly, Atilla Traverse, Madeleine Kohlberger, Timo Xu, Shawn Jamil, Fayaz Hughes, Cían Lau, Charles Chen, Justin Mahvar, Fereshteh Yatziv, Liron Chen, Tiffany Sterling, Bram Baby, Stefanie Anna Baby, Susanna Maria Lai, Jeremy Schmidgall, Samuel Yang, Lu Chen, Kejia Bjornsson, Per Reddy, Shashir Brush, Ryan Philbrick, Kenneth Asiedu, Mercy Mezerreg, Ines Hu, Howard Yang, Howard Tiwari, Richa Jansen, Sunny Singh, Preeti Liu, Yun Azizi, Shekoofeh Kamath, Aishwarya Ferret, Johan Pathak, Shreya Vieillard, Nino Merhej, Ramona Perrin, Sarah Matejovicova, Tatiana Ramé, Alexandre Riviere, Morgane Rouillard, Louis Mesnard, Thomas Cideron, Geoffrey Grill, Jean-bastien Ramos, Sabela Yvinec, Edouard Casbon, Michelle Buchatskaya, Elena Alayrac, Jean-Baptiste Lepikhin, Dmitry Feinberg, Vlad Borgeaud, Sebastian Andreev, Alek Hardin, Cassidy Dadashi, Robert Hussenot, Léonard Joulin, Armand Bachem, Olivier Matias, Yossi Chou, Katherine Hassidim, Avinatan Goel, Kavi Farabet, Clement Barral, Joelle Warkentin, Tris Shlens, Jonathon Fleet, David Cotruta, Victor Sanseviero, Omar Martins, Gus Kirk, Phoebe Rao, Anand Shetty, Shravya Steiner, David F. Kirmizibayrak, Can Pilgrim, Rory Golden, Daniel Yang, Lin Artificial Intelligence Computation and Language Computer Vision and Pattern Recognition Artificial intelligence (AI) has significant potential in healthcare applications, but its training and deployment faces challenges due to healthcare's diverse data, complex tasks, and the need to preserve privacy. Foundation models that perform well on medical tasks and require less task-specific tuning data are critical to accelerate the development of healthcare AI applications. We introduce MedGemma, a collection of medical vision-language foundation models based on Gemma 3 4B and 27B. MedGemma demonstrates advanced medical understanding and reasoning on images and text, significantly exceeding the performance of similar-sized generative models and approaching the performance of task-specific models, while maintaining the general capabilities of the Gemma 3 base models. For out-of-distribution tasks, MedGemma achieves 2.6-10% improvement on medical multimodal question answering, 15.5-18.1% improvement on chest X-ray finding classification, and 10.8% improvement on agentic evaluations compared to the base models. Fine-tuning MedGemma further improves performance in subdomains, reducing errors in electronic health record information retrieval by 50% and reaching comparable performance to existing specialized state-of-the-art methods for pneumothorax classification and histopathology patch classification. We additionally introduce MedSigLIP, a medically-tuned vision encoder derived from SigLIP. MedSigLIP powers the visual understanding capabilities of MedGemma and as an encoder achieves comparable or better performance than specialized medical image encoders. Taken together, the MedGemma collection provides a strong foundation of medical image and text capabilities, with potential to significantly accelerate medical research and development of downstream applications. The MedGemma collection, including tutorials and model weights, can be found at https://goo.gle/medgemma.
title	MedGemma Technical Report
topic	Artificial Intelligence Computation and Language Computer Vision and Pattern Recognition
url	https://arxiv.org/abs/2507.05201