U2-RadiMed
Find lesions, aid Dx — imaging you can trust
Lesion detect, Dx aid & report gen for imaging
U2-RadiMed: Find lesions, aid Dx, write reports — imaging you can trust
U2-RadiMed is Unisound's medical imaging LLM for image+text input. It helps doctors read scans, locate lesions, aid Dx, and auto-generate structured reports — an end-to-end AI assistant from image analysis to clinical decision support.
Eval
Eval covers perception (lesion detect/locate) to high-level reasoning (clinical decision & advanced medical reasoning). 18 sub-tasks span CXR / CT / MRI. U2-RadiMed ranks #1 at 55.47% overall acc., the only model above 50%.
Overall accuracy (%)
Key strengths
Key strengths
Full imaging workflow
From scan reading, lesion locate, Dx aid to structured report gen — one-stop coverage of key radiology steps.
Multimodal & multi-image understanding
Supports single/multi-image + text input. Strong at multi-sequence and temporal comparison for follow-up and complex cases.
Lesion-level visual locate
Not just a conclusion — it locates abnormalities so doctors can lock on lesions faster and reduce misses.
Clinical reasoning & decision support
Combines imaging with medical knowledge for differential Dx, treatment eval, and medical Q&A.
Medical-grade safety
AI suggestion + doctor confirm, plus result checks, keep outputs trustworthy with clear responsibility.
Tech highlights
Tech highlightsMultimodal input
Image (base64 / URL) + text instruction; single and multi-image input.
Long-context understanding
40,960-token context window for multi-image inputs and long instructions.
Structured output
Text output with bbox, Dx notes, and structured reports; supports locate results for visualization.
Use cases
Use cases
Imaging report assist
Auto-draft structured reports with findings and impressions; doctors review and edit to cut paperwork.
Screening & lesion locate
Auto-marks suspected regions to help doctors lock on faster, cut misses, and shorten read time.
Follow-up & disease tracking
Compares pre/post-treatment and multi-sequence/temporal scans to track lesion change and treatment effect.
Difficult-case differential Dx
Combines imaging features with medical knowledge to list support/exclude evidence and probability for complex cases.
Capabilities
Capabilities-
Lesion detect & locate
Detect abnormal lesions and output location (bbox); supports single-image and multi-image sequences.
-
Clinical reasoning
Reason over imaging + medical knowledge for disease tracking, multimodal differential Dx, and multimodal reports.
-
Clinical decision
Decision support on top of image understanding: sequence reading, multimodal treatment suggestions, and visual Q&A.
-
Report gen
Auto-generate diagnostic report text from images, including structured CXR and other imaging reports.
Flexible pricing, custom solutions, private deployment
Flexible billing models and dedicated customization for medical imaging scenarios, with private deployment to ensure data security and compliance
Talk to an Expert