Google releases MedGemma 1.5 4B with medical image interpretation (research.google)

Google has announced the release of MedGemma 1.5 4B, an updated version of its open medical image interpretation model, along with the debut of MedASR, a new automated speech recognition model tailored for medical settings. This enhancement marks a significant development for the AI/ML community in healthcare, as MedGemma 1.5 introduces improved capabilities for interpreting high-dimensional medical imaging modalities such as CT, MRI, and histopathology, along with enhanced text functionalities for medical documentation. The model demonstrates a marked increase in accuracy, achieving a 3% improvement in CT findings classification and up to 22% in text-based question-answering tasks compared to its predecessor. The release of these models is part of Google's broader initiative to support AI integration in healthcare through its Health AI Developer Foundations (HAI-DEF) program. Developers can leverage these tools to create innovative healthcare applications, and they will have the chance to participate in the MedGemma Impact Challenge, a hackathon with $100,000 in prizes aimed at fostering impactful AI solutions in the medical field. With the new multimodal capabilities and significant performance benchmarks—such as MedASR's 58% lower word error rate in medical dictation—Google is reinforcing its commitment to advancing AI technologies that improve clinical outcomes and decision-making processes in healthcare.