MedGemma: How does advanced AI support the monitoring of your skin moles?

Modern skin health prevention is based on two pillars: regularity and accurate documentation. In our Skin Mole Atlas, we introduce advanced support based on the MedGemma model—one of the most powerful medical AI systems in the world. Learn how Google technology helps objectively describe your moles and why the security of your data is our top priority.

 

IMPORTANT NOTICE REGARDING THE NATURE OF THE ANALYSIS

Our application and the integrated MedGemma model are not medical devices and are not intended to diagnose, detect diseases, or suggest treatment. The tool serves an educational and informational purpose—it is used to objectively describe the visual parameters of moles and facilitates their photographic documentation. The generated technical report is a supporting document that should be presented to a dermatologist during a professional dermatoscopic examination. Any skin lesion must be evaluated by a specialist.                                                                         

 

Not all artificial intelligence is “medical”

Most AI models available on the market are so-called general purpose AI models. Although they can generate text or images, their medical knowledge is often superficial. MedGemma is in a completely different category. It is a specialized multimodal model created by Google, designed from the ground up to solve clinical problems.

When choosing MedGemma for analysis in our Skin Mole Atlas, we were guided by the results of a technical report (Arxiv: 2507.05201v1), which clearly indicates the superiority of this model in tasks requiring deep medical reasoning.

Advanced structure of the MedGemma model: the system combines visual analysis with a clinical knowledge base. Source: Google Research (arXiv: 2507.05201v1).

 

Foundation of knowledge: What did MedGemma study?

The strength of an AI model depends on the quality and diversity of the data it was trained on. MedGemma has undergone a rigorous learning process, covering millions of examples from various fields of medicine. As a result, when analyzing your mole, the model uses context that simple algorithms lack.

1. A gigantic text knowledge base

Before Med-Gemma began analyzing images, it had to “pass” the most difficult medical exams. The model was trained on, among other things:

• MedQA and MedMCQA: Over 190,000 questions from American (USMLE) and Indian medical residency entrance exams.
• PubMedQA: Thousands of abstracts from peer-reviewed scientific papers, allowing the model to stay up to date with the latest academic knowledge.
• HealthSearchQA: Data on real patient questions, which helps the model communicate in a way that is understandable to the user.

2. Multimodal visual power

This is where Med-Gemma leaves the competition behind. The model analyzed hundreds of thousands of medical images to learn how to recognize pathologies:

• Histopathology (over 32 million samples!): This is a key element. The model learned from microscopic tissue samples, which allows it to understand the cellular structure of lesions, not just their external appearance.
• Dermatology (PAD-UFES-20 and internal collections): Over 50,000 precisely described images of skin lesions have enabled the model to become an expert in the visual analysis of moles and birthmarks.
• Radiology and ophthalmology: Hundreds of thousands of X-ray, CT, and MRI images have taught the model to understand anatomy and differentiate between healthy and diseased tissue.

 

How does MedGemma analyze your skin mole?

With such a wide range of training data, the analysis in our Skin Mole Atlas goes beyond a simple “color check.” The model applies clinical logic:

• Deep ABCDE analysis: Based on tens of thousands of dermatological examples, the model accurately describes asymmetry (A), borders (B), color (C), and diameter (D).
• Comparative analysis over time (Evolution): Thanks to its huge training database, MedGemma is great at picking up subtle differences in photos taken at intervals (e.g., every 6 months). It can assess whether a change in the nature of a mole is within normal limits or requires closer attention.

 

Your privacy: Model on our own resources

The MedGemma model has been deployed on our own dedicated cloud resources. What does this mean for you?

• Data isolation: Your images do not leave our secure infrastructure. No external entity has access to them.
• No “feeding” of public AI: Your data is not used to train publicly available algorithms. It is used solely to generate your private report.
• Enterprise-grade security: Independent model management allows us to fully control encryption and access to information.

 

Conscious health management

MedGemma is currently one of the best documented and safest medical AI models in the world. By implementing it in our Skin Mole Atlas, we give you a tool that combines knowledge from 32 million histopathological samples with the latest advances in deep learning.

Please note that this technology is designed to support your vigilance, not replace your doctor. Keeping a Skin Mole Atlas is a form of modern prevention that allows you to come to your dermatologist’s office with complete, reliable documentation and a history of changes in your skin.

 

Bibliography:

Google Research, MedGemma: Foundations for Health AI, Technical Report, 2024. Available online: https://arxiv.org/pdf/2507.05201v1

 

✉️ Try Medify.me and create your private digital health account: https://medify.me