Automated FDG uptake/PET-CT fused scan diagnosis of various lymph node tumors using object detection AI techniques

This study proposes an automated method for diagnosing lymph node tumors using a fused FDG uptake/PET-CT dataset and AI-based object detection. The authors created a new LN dataset by merging CT and PET images for each patient, then denoising and annotating 13 lymph node classes across body organs. Data were split 80/10/10 for training, validation, and testing, with augmentation applied only to the training set and a 5-fold cross-validation performed for reliability. The object-detection module uses a modified YOLOv8, with kernel and backbone tuned and comparisons made against eight one-stage architectures, including YOLOv7 through YOLOv12 and YOLONas. Results show improvements in precision, recall, mAP50, and DSC of 78%, 75%, 81%, and 76%, respectively.