Mehmet Ozan Unal

We propose a prompt-conditioned framework built on MedSigLIP that injects textual priors via Feature-wise Linear Modulation (FiLM) and multi-scale pooling. Text prompts condition patch-token features on clinical intent, enabling data-efficient learning and rapid adaptation. The architecture combines global, local, and texture-aware pooling through separate regression heads fused by a lightweight MLP, trained with pairwise ranking loss. Evaluated on the LDCTIQA2023 (a public LDCT quality assessment challenge) with 1,000 training images, we achieve PLCC = 0.9575, SROCC = 0.9561, and KROCC = 0.8301, surpassing the top-ranked published challenge submissions and demonstrating the effectiveness of our prompt-guided approach.

Block-Matching and 3D Filtering (BM3D) exploits non-local self-similarity priors for denoising but relies on fixed parameters. Deep models such as U-Net are more flexible but often lack interpretability and fail to generalize across noise regimes. In this study, we propose Deep Unfolded BM3D (DU-BM3D), a hybrid framework that unrolls BM3D into a trainable architecture by replacing its fixed collaborative filtering with a learnable U-Net denoiser. This preserves BM3D's non-local structural prior while enabling end-to-end optimization. We evaluate DU-BM3D on low-dose CT (LDCT) denoising and show that it outperforms classic BM3D and standalone U-Net across simulated LDCT at different noise levels, yielding higher PSNR and SSIM, especially in high-noise conditions.

Deep learning-based low-dose computed tomography reconstruction methods already achieve high performance on standard image quality metrics like peak signal-to-noise ratio and structural similarity index measure. Yet, they frequently fail to preserve the critical anatomical details needed for diagnostic tasks. This fundamental limitation hinders their clinical applicability despite their high metric scores. We propose a novel task-adaptive reconstruction framework that addresses this gap by incorporating a frozen pre-trained task network as a regularization term in the reconstruction loss function. Unlike existing joint-training approaches that simultaneously optimize both reconstruction and task networks, and risk diverging from satisfactory reconstructions, our method leverages a pre-trained task model to guide reconstruction training while still maintaining diagnostic quality. We validate our framework on a liver and liver tumor segmentation task. Our task-adaptive models achieve Dice scores up to 0.707, approaching the performance of full-dose scans (0.874), and substantially outperforming joint-training approaches (0.331) and traditional reconstruction methods (0.626). Critically, our framework can be integrated into any existing deep learning-based reconstruction model through simple loss function modification, enabling widespread adoption for task-adaptive optimization in clinical practice. Our codes are available at: https://github.com/itu-biai/task_adaptive_ct

Herein, a novel and computationally efficient near field shaping technique for nondestructive hyperthermia applications is presented. The aim of nondestructive hyperthermia studies is to treat cancer cells in the body without damaging the healthy cells using the focused near‐field characteristic of the antenna array. However, optimal near field shaping requires an optimization process for having accurately focus the energy into the desired coordinates, which is, a computational expensive and inefficient task due to the nature of full‐wave electromagnetic (EM) analysis. To enable computationally efficient, and accurate near field shape optimization, a surrogate model is required. To achieve this, the near field radiation of the designed horn antenna array in free space is obtained by using a fast and reliable technique, analytical regularization method (ARM). After that, by using Artificial Neural Network based regression algorithm, a data driven surrogate model is created using the data samples generated by ARM. By this means, near field shaping optimization for focusing the energy to the region where the tumor is located can be achieved, as a multi‐objective optimization problem to be solved via data driven surrogate model assisted optimization techniques. The performance results of K‐fold validation and hold‐out data sets for proposed data driven surrogate model are calculated using Mean Absolute Error metric which are obtained as 0.9 × 10⁻³ and 1.4 × 10⁻³ respectively. By observing the presented study case scenario, after 10 min of illumination with antenna array with optimally selected configurations, the targeted area (position of tumor) had reached to the critical temperature of 43 Celsius. Thus, it can be said that the proposed method is an efficient method, both in terms of (1) destroying tumor cells by heating without damaging the healthy cells, and (2) being a computationally efficient method.

Low-dose CT imaging requires reconstruction from noisy indirect measurements which can be defined as an ill-posed linear inverse problem. In addition to conventional FBP method in CT imaging, recent compressed sensing based methods exploit handcrafted priors which are mostly simplistic and hard to determine. More recently, deep learning (DL) based methods have become popular in medical imaging field. In CT imaging, DL based methods try to learn a function that maps low-dose images to normal-dose images. Although the results of these methods are promising, their success mostly depends on the availability of high-quality massive datasets. In this study, we proposed a method that does not require any training data or a learning process. Our method exploits such an approach that deep convolutional neural networks (CNNs) generate patterns easier than the noise, therefore randomly initialized generative neural networks can be suitable priors to be used in regularizing the reconstruction. In the experiments, the proposed method is implemented with different loss function variants. Both analytical CT phantoms and real-world CT images are used with different views. Conventional FBP method, a popular iterative method (SART), and TV regularized SART are used in the comparisons. We demonstrated that our method with different loss function variants outperforms the other methods both qualitatively and quantitatively.