Google Releases Free AI That Reads 3D Medical Scans

Doctors and hospitals around the world just gained access to AI tools that could transform how they diagnose diseases and write reports.

Google Research released MedGemma 1.5 on January 13, a free medical AI that can read full 3D CT scans and MRIs the way human specialists do. Unlike earlier versions that only viewed single images, this breakthrough interprets entire volumes at once, spotting patterns across dozens of slices that might otherwise be missed.

The improvements are dramatic. The AI jumped from 51% to 65% accuracy on MRI analysis and now generates pathology reports that match specialized systems. It can compare chest X-rays over time to track disease progression and pull structured data from messy lab reports with 78% accuracy, up from 60%.

Google paired the imaging AI with MedASR, a speech recognition system built specifically for medical dictation. General-purpose transcription tools make errors on 28% of medical words, stumbling over drug names and anatomy terms. MedASR cuts that to just 5%, an 82% reduction in mistakes.

The combination creates a seamless workflow: doctors dictate findings, the system transcribes accurately, then the AI analyzes images and helps generate reports. Both tools run on modest hardware, making them accessible even to smaller clinics.

Real hospitals are already seeing results. Malaysia's Ministry of Health uses an earlier version to help doctors navigate 150+ clinical guidelines instantly. Taiwan analyzed 30,000 pathology reports to improve lung cancer surgical planning.

The Ripple Effect

What makes this release special is the word "open." Previous medical AI this powerful was locked behind paywalls or required massive computing resources. Google released full code, model weights, and tutorials on Hugging Face, letting any developer or researcher adapt the technology for free.

Smaller hospitals in underserved regions can now deploy the same diagnostic assistance as major academic centers. Researchers can fine-tune the models for rare diseases or local populations without starting from scratch. A $100,000 Kaggle challenge is already spurring creative applications from the global developer community.

The 4-billion-parameter size keeps it efficient enough to run locally, addressing privacy concerns around patient data. Integration with standard hospital systems is built in, and Google Cloud offers ready-made deployment options for institutions that want them.

Healthcare adopts AI twice as fast as other industries because the need is urgent: radiologists facing overwhelming scan volumes, pathologists reading gigapixel slides, doctors documenting visits while trying to focus on patients. These tools don't replace human judgment but they can catch what eyes miss and give time back for actual care.

The technology is spreading fast, and the barriers just got a lot lower.