Abstract
Radiologists are tasked with interpreting a large number of images in a dailybase, with the responsibility of generating corresponding reports. Thisdemanding workload elevates the risk of human error, potentially leading totreatment delays, increased healthcare costs, revenue loss, and operationalinefficiencies. To address these challenges, we initiate a series of work ongrounded Automatic Report Generation (AutoRG), starting from the brain MRIinterpretation system, which supports the delineation of brain structures, thelocalization of anomalies, and the generation of well-organized findings. Wemake contributions from the following aspects, first, on dataset construction,we release a comprehensive dataset encompassing segmentation masks of anomalyregions and manually authored reports, termed as RadGenome-Brain MRI. This dataresource is intended to catalyze ongoing research and development in the fieldof AI-assisted report generation systems. Second, on system design, we proposeAutoRG-Brain, the first brain MRI report generation system with pixel-levelgrounded visual clues. Third, for evaluation, we conduct quantitativeassessments and human evaluations of brain structure segmentation, anomalylocalization, and report generation tasks to provide evidence of itsreliability and accuracy. This system has been integrated into real clinicalscenarios, where radiologists were instructed to write reports based on ourgenerated findings and anomaly segmentation masks. The results demonstrate thatour system enhances the report-writing skills of junior doctors, aligning theirperformance more closely with senior doctors, thereby boosting overallproductivity.