1. Главная
  2. Ресурсы
  3. Библиотечный поиск
Dynamic PETReconstructionUtilizing aSpatiotemporal 4DDe-noising Kernel Ju-Chieh ChengConnor BevingtonArman Rahmim2018 год

Dynamic PET Reconstruction Utilizing a Spatiotemporal 4D De-noising Kernel
статья

Полный текстСтраница публикацииПубликация в OpenAlex
Аннотация: We propose novel 4D de-noised image reconstruction frame works, followed by extensive validation using 4D simulations, experimental phantom as well as clinical patient data. Previously, it was demonstrated that our 3D de-noised reconstruction, which applies the HighlY constrained backPRojection (HYPR) de-noising operator After each Update of OSEM (HYPR-AU-OSEM), can achieve noise reduction and improve the reproducibility in contrast recovery without degrading accuracy in terms of resolution and contrast for single frame reconstruction. Moreover, the method does not require any prior information and is not computationally intensive. In this work, we propose the 4D extension of HYPR-AU-OSEM (i.e. HYPR4D-AU-OSEM) for dynamic imaging. Further, we incorporate the proposed 4D de-noising operator within the recently proposed kernelized reconstruction frame work (i.e. HYPR4D-K-OSEM) inspired by machine learning. In short, the proposed methods make use of the spatiotemporal high frequency features extracted from the 4D composite, generated directly within the reconstruction, to preserve the 4D resolution and constrain the noise increment in both spatial and temporal domains. Results from the simulations, experimental phantom, and patient data showed that the proposed methods outperformed the standard OSEM with post filter in terms of 4D resolution, contrast recovery coefficient vs noise trade-off, and accuracy in time-activity-curves (TAC) and binding potential (BP ND ) values. In particular, the root mean squared error in regional BP ND values was reduced from ~8% to ~3% using the proposed methods. Compared to the conventional 3D composite, the 4D composite achieved 50% lower mean absolute error in TACs. Comparable results were obtained between AU and kernel methods. In summary, the improvement in 4D resolution and noise reduction obtained from the proposed methods can produce more robust and accurate image features without any prior information, as compared to the conventional methods.
Год издания: 2018
Авторы: Ju-Chieh Cheng, Connor Bevington, Arman Rahmim, Ivan S. Klyuzhin, Julian C. Matthews, Ronald Boellaard, Vesna Sossi
Ключевые слова: Medical Imaging Techniques and Applications, Advanced MRI Techniques and Applications, Advanced X-ray Imaging Techniques
Другие ссылки: doi.org (HTML)
University of Groningen research database (University of Groningen / Centre for Information Technology) (HTML)
University of Groningen research database (University of Groningen / Centre for Information Technology) (PDF)
University of Groningen research database (University of Groningen / Centre for Information Technology) (HTML)