arXiv:2510.27680v2 Announce Type: replace-cross 
Abstract: Generating automated reports for 3D positron emission tomography (PET) is an important and challenging task in medical imaging. PET plays a vital role in oncology, but automating report generation is difficult due to the complexity of whole-body 3D volumes, the wide range of potential clinical findings, and the limited availability of annotated datasets. To address these challenges, we introduce PETARSeg-11K, the first large-scale, publicly available dataset that provides lesion-level correspondence between 3D PET/CT volumes and free-text radiological findings. It comprises 11,356 lesion descriptions paired with 3D segmentations. Second, we propose PETAR-4B, a 3D vision-language model designed for mask-aware, spatially grounded PET/CT reporting. PETAR-4B jointly encodes PET, CT, and 3D lesion segmentation masks, using a 3D focal prompt to capture fine-grained details of lesions that normally comprise less than 0.1% of the volume. Evaluations using automated metrics show PETAR-4B substantially outperforming all 2D and 3D baselines. A human study involving five physicians -- the first of its kind for automated PET reporting -- confirms the model's clinical utility and establishes correlations between automated metrics and expert judgment. This work provides a foundational dataset and a novel architecture, advancing 3D medical vision-language understanding in PET.

يمثل تقديم PETARSeg-11K، وهو مجموعة بيانات واسعة النطاق، وPETAR-4B، وهو نموذج للرؤية واللغة ثلاثي الأبعاد، تقدمًا كبيرًا في أتمتة توليد التقارير لتصوير الانبعاثات الموجبة (PET) وتصوير الأشعة المقطعية (CT). تهدف هذه المبادرة إلى معالجة تعقيدات التصوير الجسدي الكامل ونقص مجموعات البيانات المعلّمة، بهدف تحسين دقة وكفاءة التقارير الإشعاعية.

La introducción de PETARSeg-11K, un conjunto de datos a gran escala, y de PETAR-4B, un modelo de visión-lenguaje 3D, marca un avance significativo en la automatización de la generación de informes para la tomografía por emisión de positrones (PET) y la imagenología CT. Esta iniciativa aborda las complejidades de la imagenología corporal completa y la escasez de conjuntos de datos anotados, con el objetivo de mejorar la precisión y la eficiencia de los informes radiológicos.

L'introduction de PETARSeg-11K, un ensemble de données à grande échelle, et de PETAR-4B, un modèle de vision-langage 3D, marque une avancée significative dans l'automatisation de la génération de rapports pour la tomographie par émission de positons (PET) et l'imagerie CT. Cette initiative vise à résoudre les complexités de l'imagerie corporelle entière et la rareté des ensembles de données annotées, dans le but d'améliorer la précision et l'efficacité des rapports radiologiques.

The introduction of PETARSeg-11K, a large-scale dataset, and PETAR-4B, a 3D vision-language model, marks a significant advancement in automating report generation for 3D positron emission tomography (PET) and CT imaging. This initiative addresses the complexities of whole-body imaging and the scarcity of annotated datasets, aiming to enhance the accuracy and efficiency of radiological reporting.

PETAR: Localized Findings Generation with Mask-Aware Vision-Language Modeling for PET Automated Reporting

Was this article worth reading? Share it

GPTHumanizer

Airparser

Lovart-Al Ip Avatar Design