English

A Novel Loss Function Incorporating Imaging Acquisition Physics for PET Attenuation Map Generation using Deep Learning

Image and Video Processing 2019-09-05 v1 Computer Vision and Pattern Recognition Machine Learning Medical Physics

Abstract

In PET/CT imaging, CT is used for PET attenuation correction (AC). Mismatch between CT and PET due to patient body motion results in AC artifacts. In addition, artifact caused by metal, beam-hardening and count-starving in CT itself also introduces inaccurate AC for PET. Maximum likelihood reconstruction of activity and attenuation (MLAA) was proposed to solve those issues by simultaneously reconstructing tracer activity (λ\lambda-MLAA) and attenuation map (μ\mu-MLAA) based on the PET raw data only. However, μ\mu-MLAA suffers from high noise and λ\lambda-MLAA suffers from large bias as compared to the reconstruction using the CT-based attenuation map (μ\mu-CT). Recently, a convolutional neural network (CNN) was applied to predict the CT attenuation map (μ\mu-CNN) from λ\lambda-MLAA and μ\mu-MLAA, in which an image-domain loss (IM-loss) function between the μ\mu-CNN and the ground truth μ\mu-CT was used. However, IM-loss does not directly measure the AC errors according to the PET attenuation physics, where the line-integral projection of the attenuation map (μ\mu) along the path of the two annihilation events, instead of the μ\mu itself, is used for AC. Therefore, a network trained with the IM-loss may yield suboptimal performance in the μ\mu generation. Here, we propose a novel line-integral projection loss (LIP-loss) function that incorporates the PET attenuation physics for μ\mu generation. Eighty training and twenty testing datasets of whole-body 18F-FDG PET and paired ground truth μ\mu-CT were used. Quantitative evaluations showed that the model trained with the additional LIP-loss was able to significantly outperform the model trained solely based on the IM-loss function.

Keywords

Cite

@article{arxiv.1909.01394,
  title  = {A Novel Loss Function Incorporating Imaging Acquisition Physics for PET Attenuation Map Generation using Deep Learning},
  author = {Luyao Shi and John A. Onofrey and Enette Mae Revilla and Takuya Toyonaga and David Menard and Jo-seph Ankrah and Richard E. Carson and Chi Liu and Yihuan Lu},
  journal= {arXiv preprint arXiv:1909.01394},
  year   = {2019}
}

Comments

Accepted at MICCAI 2019

R2 v1 2026-06-23T11:04:31.860Z