PET2Rep: Towards Vision-Language Model-Drived Automated Radiology Report Generation for Positron Emission Tomography

The introduction of PET2Rep marks a pivotal advancement in the automation of radiology report generation for positron emission tomography (PET), a vital imaging technique in oncology and neurology. Traditional report creation is labor-intensive and time-consuming, which can hinder clinical decision-making. Recent developments in vision-language models (VLMs) have shown promise in medical applications, yet their use in PET imaging has been limited. PET2Rep addresses this gap by providing a large-scale benchmark dataset that uniquely captures whole-body image-report pairs with metabolic information. This dataset not only facilitates the evaluation of VLMs in generating accurate and informative reports but also introduces new clinical efficacy metrics to assess the quality of radiotracer uptake descriptions in key organs. By bridging the existing gaps in PET imaging resources, PET2Rep is set to enhance the efficiency and effectiveness of radiology practices.