Victoria Bazir
⬤ Google's newest release, MedGemma 1.5, marks a serious leap forward for open medical AI models targeting healthcare developers. The big news? MedGemma now handles way more than basic 2D medical images—it's packing full support for higher-dimensional imaging, including complete 3D CT and MRI volumes. The workflow shows exactly how developers can take MedGemma from initial use-case planning and model selection all the way through evaluation, fine-tuning, and full-scale deployment.
⬤ The standout feature in MedGemma 1.5 is how it processes entire imaging studies instead of just looking at individual slices. Developers can feed it multiple CT or MRI slices, or bundles of pathology image patches, along with text prompts—and the model analyzes everything as one complete scan. There's also beefed-up longitudinal imaging support now, letting you compare studies across different time points. Think tracking disease progression through several chest X-rays instead of making judgments based on a single snapshot.
⬤ MedGemma 1.5 brings anatomical localization to chest X-rays, pinpointing where specific structures sit and measuring spatial accuracy using intersection-over-union metrics. Medical document processing got a boost too—the model pulls structured data from lab reports and clinical documents more accurately, with higher macro F1 scores showing better precision and recall across the board. The updated 4B parameter model also crushes text and medical record tasks compared to the earlier 4B version.
⬤ Beyond imaging, MedGemma 1.5 works hand-in-hand with MedASR, Google's medical speech-to-text tool. On chest X-ray dictation tasks, MedASR hit a 5.2% word error rate—way better than Whisper large-v3's 12.5%. Sundar Pichai called MedGemma 1.5 a major upgrade, emphasizing how it natively handles full 3D scans with serious efficiency. This release shows Google's commitment to building open, multimodal AI infrastructure for healthcare—setting the stage for broader cloud adoption and next-level digital health innovation.
Victoria Bazir
