Anatomical-guided attention enhances unsupervised PET image denoising performance

•We propose an unsupervised PET image denoising with anatomical-guided attention that combines two subnetworks: encoder-decoder and deep decoder subnetworks.•Encoder-decoder subnetwork provides both multi-scale spatial and semantic features to deep decoder subnetwork using attention gates.•The proposed method demonstrates state-of-the-art performance compared to other denoising methods.•Our results also provide useful insights in unsupervised CNN-based restoration optimization process.