The report introduces Harrison.Rad 1.5 (HR1.5), a radiology-specific multimodal language model that combines interleaved text and images with prior studies and clinical context to draft structured or unstructured reports. It covers computed radiography, chest, musculoskeletal, abdominal, spine, and pelvic x-rays, as well as mammography. Training uses radiology-report domain adaptation, contrastive vision-encoder training on approximately 6 million image-report instances with curriculum-based hard negatives, and multi-turn visual-question-answering fine-tuning. The authors evaluate it on RadBench, a simulated FRCR 2B Short Case examination, ReXGradient, and internal multi-modality datasets.
No heat snapshots are available in the last 24 hours.