Image Recognition
Imaging protocol and image reconstructionExaminations were performed using a latest generation six-ring digital detector PET/CT scanner (Discovery MI Gen 2, GE Healthcare, Waukesha, WI). The SNR served as a surrogate for objective image quality. Automated image quality assessment using machine learningFor automated image quality assessment, the fast.ai deep learning library14 was used in conjunction with a Res-Net-3415, a 34-layer residual convolutional network pre-trained on the Image-Net dataset (https://arxiv.org/abs/1512.03385). The MIP datasets from PET images were exported with a total of 400 DICOM files. Assessment of group differences was determined using an unpaired t-test after ensuring a normal distribution of the data using the Shapiro–Wilk test.