Evaluating Hybrid Deep Learning and Traditional Methods for PET Image Reconstruction: A Comparative Study
Main Article Content
Keywords
Positron Emission Tomography, Deep Learning, Generative Adversarial Networks, Image Reconstruction, Evaluation
Abstract
Background/Objectives: Positron Emission Tomography (PET) images typically exhibit high noise levels and limited spatial resolution. This paper presents a comparative investigation of traditional PET image reconstruction methods, including Filtered Back Projection (FBP), Algebraic Reconstruction Technique (ART), and Ordered Subset Expectation Maximization (OSEM), alongside hybrid approaches that incorporate deep learning techniques.
Methods: The deep learning approach employed in this work is based on Generative Adversarial Networks (GANs), a powerful framework well suited for inverse problems and image generation tasks such as PET reconstruction. This approach is tested on a publicly available dataset consisting of PET images stored in DICOM format. Performance is evaluated using two standard metrics: the Peak Signal-to-Noise Ratio (PSNR) and the Mean Squared Error (MSE).
Results: The results demonstrate that our proposed methods outperform existing approaches in terms of performance while requiring less reconstruction time. Quantitatively, the Peak Signal-to-Noise Ratio (PSNR) of the reconstructed images is approximately 50 dB. Qualitatively, the observed high image quality supports these quantitative findings.
Conclusions: Our proposed hybrid method is highly effective for noisy PET images, enabling accurate reconstruction and preserving pertinent information and regions of interest, thereby facilitating medical diagnosis.
Downloads
Abstract 30 | PDF Downloads 3
References
2.Hashimoto, F., Ote, K., Onishi, Y. PET image reconstruction incorporating deep image prior and a forward projection model. IEEE Trans Radiat. 2022;6(8):841–846. https://doi.org/10.1109/TRPMS.2022.3161569.
3.Gong, K., Guan, J., Kim, K., et al. Iterative PET image reconstruction using convolutional neural network representation. IEEE Trans Med Imaging. 2019;38(3):675–685. https://doi.org/10.1109/TMI.2018.2869871.
4.Yi, X., Walia, E., Babyn, P.S. Generative adversarial network in medical imaging: a review. Med Image Anal. 2019;58(2):101552. https://doi.org/10.1016/j.media.2019.101552.
5.Andersen, A.H. and Kak, A.C. Simultaneous algebraic reconstruction technique (SART): a superior implementation of the ART algorithm. Ultrason Imaging. 1984;6(1):81–94. https://doi.org/10.1016/0161-7346(84)90008-7.
6.Long, Y., Huo, X., Liu, H., et al. An extended simultaneous algebraic reconstruction technique for imaging the ionosphere using GNSS data and its preliminary results. Remote Sens. 2023;15(11):2939. https://doi.org/10.3390/rs15112939.
7.Häggström, I., Schmidtlein, C.R., Campanella, G., et al. DeepPET: a deep encoder-decoder network for directly solving the PET image reconstruction inverse problem. Med Image Anal. 2019;54:253–262. https://doi.org/10.1016/j.media.2019.03.013.
8.Parkinson’s Progression Markers Initiative (PPMI). Available online: https://www.ppmi-info.org/.
9.ARAMIS Project-Team. Algorithms, models and methods for images and signals of the human brain. Inria Paris Centre, CNRS, INSERM, Sorbonne Université, France, 2023. Available online: https://radar.inria.fr/rapportsactivite/RA2023/aramis/ARAMIS-RA-2023.pdf.
10.Wang, Z., She, Q., Ward, T.E. Generative adversarial networks in computer vision: a survey and taxonomy. Preprint. 2019, arXiv:1906.01529. Available online: https://arxiv.org/abs/1906.01529.
11.Beckmann, M. and Nickel, J. Optimized filter functions for filtered back projection reconstructions. https://doi.org/10.3934/ipi.2025003.
12.Hartling, K., Mahoney, F., Rand, E.T., et al. A comparison of algebraic reconstruction techniques for a single-detector muon computed tomography system. Nucl Instrum Methods Phys Res A. 2021;987:164834. https://doi.org/10.1016/j.nima.2020.164834.
13.Zeng, T., Gao, J., Gao, D. A GPU-accelerated fully 3D OSEM image reconstruction for a high-resolution small animal PET scanner using dual-ended readout detectors. Phys Med Biol. 2020;65(24). https://doi.org/10.1088/1361-6560/aba6f9.
14.Anaya-Sánchez, H., Altamirano Robles L., Díaz Hernández, R., et al. WGAN-GP for synthetic retinal image generation: enhancing sensor-based medical imaging for classification models. Sensors. 2024;25(1):167. https://doi.org/10.3390/s25010167.
15.Al-Najjar, Y. Comparative analysis of image quality assessment metrics: MSE, PSNR, SSIM and FSIM. Int J Sci Res. 2024;13(3):110–114. https://doi.org/10.21275/SR24302013533.
16.Hashimoto, F., Onishi, Y., Ote, K., et al. Deep learning-based PET image denoising and reconstruction: a review. Radiol Phys Technol. 2024;17:24–46. https://doi.org/10.1007/s12194-024-00780-3.
17.He, W., Zhao, Y., Zhao, X., et al. A CNN-based four-layer DOI encoding detector using LYSO and BGO scintillators for small animal PET imaging. Phys Med Biol. 2023;68(9). https://doi.org/10.1088/1361-6560/accc07.
Article Sidebar
Article Details
This work is licensed under a Creative Commons Attribution 4.0 International License.
Author(s) must obtain all parties consent [co-author(s), others if applicable] and submit the acceptance of Copyright Agreement with their paper. Written permission must be obtained by the author for material that has been published in copyrighted material; this includes tables, figures, and quoted text that exceeds 150 words. A copy of all permissions must accompany the manuscript when published in copyrighted material. The author(s) hereby represents and warrants that the paper is original and that he/she is the author of the paper, except for material that is clearly identified as to its original source, with permission notices from the copyright owners where required. Author(s) must clearly indicate that approval for publication has been received in cases of institutional ownership.
In all submitted material authors retain all copyrights; however, the GlobalCE Journal reserve the right to reprint all or portions of the article and to post all or part of the article online. GlobalCE Journal reserves the right to edit manuscripts as required to publish in the journal. Authors are responsible for obtaining any and all clearances as appropriate. Unless stated otherwise, all articles are open-access and distributed under the terms of the Creative Commons Attribution License (CC BY 4.0) which permits unrestricted use, distribution, and reproduction in any medium, provided the original author(s) and original work (e.g. first published in the Global Clinical Engineering Journal) is properly cited with the original URL and bibliographic citation information. The complete bibliographic information, a link to the original publication on www.Globalce.org well as the copyright and license information must be included. No use, distribution or reproduction is permitted which does not comply with these terms.
