In the field of radiology, the demand for imaging is growing faster than the expansion of the radiology workforce, leading to reporting backlogs that cannot be resolved through training and recruitment alone. The most direct opportunity lies in reducing the time and effort radiologists spend on report generation, a task that involves interpreting images, integrating clinical history and prior studies, and drafting structured findings.
We present Harrison.Rad 1.5 (HR1.5), a radiology-specific multimodal large language model that accepts interleaved text and visual inputs, generating structured and unstructured text across plain-film radiology, including computed radiography, chest, musculoskeletal, abdominal, spine, pelvic X-rays, and mammography.
HR1.5 is trained through a three-stage pipeline: domain adaptation of a base language model on radiology reports, contrastive vision-encoder training with curriculum-based hard negatives on approximately 6 million image-report instances, and visual-question-answering fine-tuning on multi-turn conversations.
We evaluate it using a Findings-Diagnosis scoring framework that extends RadGraph-XL entity extraction with ontology-based synonym matching and polarity-contradiction detection, benchmarked on RadBench, a simulated FRCR 2B Short Case examination scored against Angoff-method thresholds. HR1.5 is the only system evaluated to meet the simulated FRCR passing standard, achieving the highest accuracy on closed-format clinical questions across anatomical regions, on internal multi-body-part and mammography reporting, and on the primary clinically-aligned score for public chest reporting.
Additionally, we examine explainability and model behavior, including question-sensitive Grad-CAM heatmaps, attention analysis, and confidence estimation, to support responsible future evaluation toward clinical use, and a framework for clinically grounded assessment of report quality.
Blogger's Review: HR1.5 demonstrates the immense potential of multimodal large language models in radiology, significantly alleviating the burden on radiologists through automated report generation, thereby enhancing efficiency. Its three-stage training pipeline and advanced evaluation framework lay a solid foundation for clinical applications, promising to drive the intelligent evolution of radiology.