A Systematic Comparison of Deep Learning Methods for Lymphoma Segmentation in PET/CT Images

Sammanfattning

This research project systematically studies the PET/CT segmentation task in the lymphoma diagnosis scenario. By establishing a complete output processing, model training, evaluation, and visualization process based on the AutoPet dataset, the entire research project systematically analyzes and compares the performance of the three models, Unet, SegResnet, and SegResnetDS, in the lymphoma segmentation scenario. In addition, this study also experimentally compares the effect of training the model on PET and CT data separately and training it together. By establishing a complete and systematic evaluation index, the conclusions obtained in this research project can provide important data references for practitioners and engineers in related fields.