Browsing by Subject "Plug and play"

Now showing 1 - 2 of 2

Open Access
Deep learning reconstruction for single pixel imaging with generative adversarial networks
(IEEE, 2023-09-11) Güven, Baturalp; Güngör, A.; Bahçeci, M. U.; Çukur, Tolga
Single pixel imaging (SPI) enables high-resolution imaging through multiple coded measurements based on low-resolution snapshots. An inverse problem can then be solved to reconstruct a high-resolution image given the coded measurements. There has been recent interest in adoption of deep neural networks in SPI reconstruction. However, existing methods are commonly trained with pixel-wise loss terms such as the ℓ 1 -norm loss, which can result in spatial blurring and poor sensitivity to structural details. In this study, we propose a novel approach for deep SPI reconstruction based on an unrolled conditional generative adversarial network (cGAN) model. The generator estimates the high-resolution image using coded low-resolution measurements by iterating across a cascade of denoising and data-consistency modules. Meanwhile, the discriminator distinguishes real versus synthesized high-resolution images. The architecture is trained end-to-end via a combined pixel-wise and adversarial loss to enhance sensitivity to structural details. The proposed method is demonstrated against existing SPI reconstruction methods, and ablation studies are performed to demonstrate the individual model components. The proposed method outperforms competing methods in terms of both quantitative metrics and visual quality.
Open Access
PP-MPI: A deep plug-and-play prior for magnetic particle imaging reconstruction
(Springer Cham, 2022-09) Aşkın, Barış; Güngör, Alper; Alptekin Soydan, D.; Sarıtaş, Emine Ülkü; Top, C. B.; Çukur, Tolga; Haq, Nandinee; Maier, Andreas; Qin, Chen; Johnson, Patricia; Würfl, Tobias; Yoo, Jaejun
Magnetic particle imaging (MPI) is a recent modality that enables high contrast and frame-rate imaging of the magnetic nanoparticle (MNP) distribution. Based on a measured system matrix, MPI reconstruction can be cast as an inverse problem that is commonly solved via regularized iterative optimization. Yet, hand-crafted regularization terms can elicit suboptimal performance. Here, we propose a novel MPI reconstruction “PP-MPI” based on a deep plug-and-play (PP) prior embedded in a model-based iterative optimization. We propose to pre-train the PP prior based on a residual dense convolutional neural network (CNN) on an MPI-friendly dataset derived from magnetic resonance angiograms. The PP prior is then embedded into an alternating direction method of multiplier (ADMM) optimizer for reconstruction. A fast implementation is devised for 3D image reconstruction by fusing the predictions from 2D priors in separate rectilinear orientations. Our demonstrations show that PP-MPI outperforms state-of-the-art iterative techniques with hand-crafted regularizers on both simulated and experimental data. In particular, PP-MPI achieves on average 3.10 dB higher peak signal-to-noise ratio than the top-performing baseline under variable noise levels, and can process 12 frames/sec to permit real-time 3D imaging.