PET Partial Volume Correction Based on Unsupervised Deep Residual Compensation Model

Jiankai Wu, Zhongxue Wu, Jianzhong He, Qingrun Zeng, Zan Chen, Yuanjing Feng, Jianan Cui

Abstract: Partial volume effect (PVE) in PET imaging arises due to the limited spatial resolution, which can lead to quantitative biases. Our study aims to overcome the adverse impact of PVE on PET images through an unsupervised deep residual compensation model with anatomical prior. The proposed partial volume correction (PVC) method first predicts an initial blur kernel for the PVE-affected PET image from a random Gaussian blur kernel. Then, a deep residual compensation network is introduced to compensate for the error caused by inaccurate blur kernel prediction. The whole model is unsupervised which only needs single patient's PET image as the training label. The corresponding MR image works as the network input to provide anatomical information. Experiments on simulation and real clinical brain datasets show that our method can effectively recover fine details in PET images and improve the accuracy of quantitative analysis of PET images.