CT image reconstruction based on sparse representation an‎d deep Networks

10/11/2025 12:00:00 AM

Computed Tomography (CT) is one of the most widely used diagnostic methods in medicine, but the high radiation dose in this type of imaging has always raised serious concerns regarding patient health. One approach to dose reduction is using limited-view imaging, which faces the challenge of reconstructing images with acceptable quality. In this research, an innovative algorithm called PnP-AugHQS was designed an‎d developed for CT image reconstruction with limited data. This algorithm, by leveraging the Plug-an‎d-Play (PnP) framework, utilizing the deep denoiser DRUNet, an‎d adaptive Half-Quadratic Splitting (HQS) optimization along with dynamic adjustment mechanisms for key parameters, successfully established a balance between speed, accuracy, an‎d stability. To eva‎luate the performance of the proposed algorithm, extensive experiments were conducted on simulated data, an‎d the results were compared with classical reconstruction methods an‎d deep learning algorithms. The findings showed that PnP-AugHQS achieved an average improvement of 2.3 dB in PSNR compared to classical reference methods an‎d significantly reduced reconstruction time (between 120 to 300 times). Furthermore, the reconstruction quality under limited-data conditions reached a competitive level with deep learning-based methods, while not requiring specific training on CT data. Overall, the results of this research indicate that the PnP-AugHQS algorithm can be used as an efficient an‎d flexible method for limited-view CT image reconstruction. This achievement not only represents an effective step toward reducing patient radiation dose but also provides a suitable foundation for future developments in novel areas such as perfusion CT an‎d real-time reconstruction.

،مقطع‌نگاري ‌كامپيوتري، سينوگرام، بازسازي تصوير، روش نصب و اجرا(PnP)، يادگيري عميق،

،Computed Tomography, Sinogram, Image Reconstruction, Plug-and-Play (PnP), Deep Learning،

Nafiseh Zavarehzadegan

Abozar Ghafari