CyberSec Research

Magnetic Particle Imaging (MPI) is a tomographic imaging modality capable of real-time, high-sensitivity mapping of superparamagnetic iron oxide nanoparticles. Model-based image reconstruction provides an alternative to conventional methods that rely on a measured system matrix, eliminating the need for laborious calibration measurements. Nevertheless, model-based approaches must account for the complexities of the imaging chain to maintain high image quality. A recently proposed direct reconstruction method leverages weighted Chebyshev polynomials in the frequency domain, removing the need for a simulated system matrix. However, the underlying model neglects key physical effects, such as nanoparticle anisotropy, leading to distortions in reconstructed images. To mitigate these artifacts, an adapted direct Chebyshev reconstruction (DCR) method incorporates a spatially variant deconvolution step, significantly improving reconstruction accuracy at the cost of increased computational demands. In this work, we evaluate the adapted DCR on six experimental phantoms, demonstrating enhanced reconstruction quality in real measurements and achieving image fidelity comparable to or exceeding that of simulated system matrix reconstruction. Furthermore, we introduce an efficient approximation for the spatially variable deconvolution, reducing both runtime and memory consumption while maintaining accuracy. This method achieves computational complexity of O(N log N ), making it particularly beneficial for high-resolution and three-dimensional imaging. Our results highlight the potential of the adapted DCR approach for improving model-based MPI reconstruction in practical applications.

Photoacoustic imaging (PAI) is rapidly moving from the laboratory to the clinic, increasing the need to understand confounders which might adversely affect patient care. Over the past five years, landmark studies have shown the clinical utility of PAI, leading to regulatory approval of several devices. In this article, we describe the various causes of artifacts in PAI, providing schematic overviews and practical examples, simulated as well as experimental. This work serves two purposes: (1) educating clinical users to identify artifacts, understand their causes, and assess whether their impact, and (2) providing a reference of the limitations of current systems for those working to improve them. We explain how two aspects of PAI systems lead to artifacts: their inability to measure complete data sets, and embedded assumptions during reconstruction. We describe the physics underlying PAI, and propose a classification of the artifacts. The paper concludes by discussing possible advanced mitigation strategies.

Accurate three-dimensional (3D) reconstruction of cardiac chamber motion from time-resolved medical imaging modalities is of growing interest in both the clinical and biomechanical fields. Despite recent advancement, the cardiac motion reconstruction process remains complex and prone to uncertainties. Moreover, traditional assessments often focus on static comparisons, lacking assurances of dynamic consistency and physical relevance. This work introduces a novel paradigm of flow-compatible motion reconstruction, integrating anatomical imaging with flow data to ensure adherence to fundamental physical principles, such as mass and momentum conservation. The approach is demonstrated in the context of right ventricular motion, utilizing diffeomorphic mappings and multi-slice MRI to achieve dynamically consistent and physically robust reconstructions. Results show that enforcing flow compatibility within the reconstruction process is feasible and enhances the physical realism of the resulting kinematics.

Background: Adnexal masses are heterogeneous and have varied sonographic presentations, making them difficult to diagnose correctly. Purpose: Our study aimed to develop an innovative hybrid artificial intelligence/computer aided diagnosis (AI/CADx)-based pipeline to distinguish between benign and malignant adnexal masses on ultrasound imaging based upon automatic segmentation and echogenic-based classification. Methods: The retrospective study was conducted on a consecutive dataset of patients with an adnexal mass. There was one image per mass. Mass borders were segmented from the background via a supervised U-net algorithm. Masses were spatially subdivided automatically into their hypo- and hyper-echogenic components by a physics-driven unsupervised clustering algorithm. The dataset was separated by patient into a training/validation set (95 masses; 70%) and an independent held-out test set (41 masses; 30%). Eight component-based radiomic features plus a binary measure of the presence or absence of solid components were used to train a linear discriminant analysis classifier to distinguish between malignant and benign masses. Classification performance was evaluated using the area under the receiver operating characteristic curve (AUC), along with sensitivity, specificity, negative predictive value, positive predictive value, and accuracy at target 95% sensitivity. Results: The cohort included 133 patients with 136 adnexal masses. In distinguishing between malignant and benign masses, the pipeline achieved an AUC of 0.90 [0.84, 0.95] on the training/validation set and 0.93 [0.83, 0.98] on the independent test set. Strong diagnostic performance was observed at the target 95% sensitivity. Conclusions: A novel hybrid AI/CADx echogenic components-based ultrasound imaging pipeline can distinguish between malignant and benign adnexal masses with strong diagnostic performance.

This study proposes a novel imaging and reconstruction framework for dual-energy cone-beam CT (DECBCT) using only two orthogonal X-ray projections at different energy levels (2V-DECBCT). The goal is to enable fast and low-dose DE volumetric imaging with high spectral fidelity and structural accuracy, suitable for DECBCT-guided radiation therapy. We introduce a framework for 2V-DECBCT based on physics-informed dual-domain diffusion models. A cycle-domain training strategy is employed to enforce consistency between projection and volume reconstructions through a differentiable physics-informed module. Furthermore, a spectral-consistency loss is introduced to preserve inter-energy contrast during the generative process. The model is trained and evaluated using 4D XCAT phantom data under realistic anatomical motion. The method produces high-fidelity DECBCT volumes from only two views, accurately preserving anatomical boundaries and suppressing artifacts. Subtraction maps computed from the reconstructed energy volumes show strong visual and numerical agreement with ground truth. This work presents the first diffusion model-based framework for 2V-DECBCT reconstruction, demonstrating accurate structural and spectral recovery from extremely sparse inputs.

Kidney stones can cause severe pain and complications such as chronic kidney disease or kidney failure. Retrograde intrarenal surgery (RIRS), which uses laser lithotripsy to fragment stones for removal via a ureteroscope, is widely adopted due to its safety and effectiveness. However, conventional fragment removal methods using basketing and vacuum-assisted aspiration are inefficient, as they can capture only 1 to 3 fragments (1--3\,mm in size) per pass, often requiring dozens to hundreds of ureteroscope passes during a single procedure to completely remove the fragments. These limitations lead to prolonged procedures and residual fragments that contribute to high recurrence rates. To address these limitations, we present a novel spinner device that enables ultra-efficient fragment removal through spinning-induced localized suction. The spinner generates a three-dimensional spiral and circulating flow field that dislodges and draws fragments into its cavity even from distances over 20\,mm, eliminating the need to chase fragments. It can capture over 60 fragments (0.5--2\,mm) or over 15 larger fragments (2--3\,mm) in a single pass, significantly improving removal efficiency. In this work, the spinner design is optimized via computational fluid dynamics to maximize suction performance. \textit{In vitro} testing demonstrates near 100\% capture rates for up to 60 fragments in a single operation and superior large-distance capture efficacy compared to vacuum-assisted methods. \textit{Ex vivo} testing of the integrated spinner-ureteroscope system in a porcine kidney confirmed its high performance by capturing 45 fragments in just 4 seconds during a single pass and achieving complete fragment clearance within a few passes.

The CMS quality measure for CT mandates compliance with dose and image quality across 18 exam categories, yet the method for calculating global noise (GN) remains undefined, referencing only the Duke and Wisconsin methods. This ambiguity raises concerns about standardization and clinical applicability. We retrospectively analyzed 719 CT exams across seven protocols, comparing five GN metrics derived from the Duke and Wisconsin approaches. Results showed significant variability across metrics and protocols, with Wisconsin_tissue_mean consistently yielding the highest values. Correlation strength varied by protocol, and mode-based metrics demonstrated higher agreement, suggesting dose dependence. In the Chest PE protocol, GN values differed significantly based on reconstruction kernel. These findings underscore the need for a standardized, clinically meaningful GN metric to ensure consistent image quality evaluation and effective implementation of the CMS measure.

Understanding the mechanical behavior of brain tissue is crucial for advancing both fundamental neuroscience and clinical applications. Yet, accurately measuring these properties remains challenging due to the brain unique mechanical attributes and complex anatomical structures. This review provides a comprehensive overview of commonly used techniques for characterizing brain tissue mechanical properties, covering both invasive methods such as atomic force microscopy, indentation, axial mechanical testing, and oscillatory shear testing and noninvasive approaches like magnetic resonance elastography and ultrasound elastography. Each technique is evaluated in terms of working principles, applicability, representative studies, and experimental limitations. We further summarize existing publications that have used these techniques to measure human brain tissue mechanical properties. With a primary focus on invasive studies, we systematically compare their sample preparation, testing conditions, reported mechanical parameters, and modeling strategies. Key sensitivity factors influencing testing outcomes (e.g., sample size, anatomical location, strain rate, temperature, conditioning, and post-mortem interval) are also discussed. Additionally, selected noninvasive studies are reviewed to assess their potential for in vivo characterization. A comparative discussion between invasive and noninvasive methods, as well as in vivo versus ex vivo testing, is included. This review aims to offer practical guidance for researchers and clinicians in selecting appropriate mechanical testing approaches and contributes a curated dataset to support constitutive modeling of human brain tissue.

Purpose: Accurately classifying tissue margins during cancer surgeries is crucial for ensuring complete tumor removal. Rapid Evaporative Ionization Mass Spectrometry (REIMS), a tool for real-time intraoperative margin assessment, generates spectra that require machine learning models to support clinical decision-making. However, the scarcity of labeled data in surgical contexts presents a significant challenge. This study is the first to develop a foundation model tailored specifically for REIMS data, addressing this limitation and advancing real-time surgical margin assessment. Methods: We propose FACT, a Foundation model for Assessing Cancer Tissue margins. FACT is an adaptation of a foundation model originally designed for text-audio association, pretrained using our proposed supervised contrastive approach based on triplet loss. An ablation study is performed to compare our proposed model against other models and pretraining methods. Results: Our proposed model significantly improves the classification performance, achieving state-of-the-art performance with an AUROC of $82.4\% \pm 0.8$. The results demonstrate the advantage of our proposed pretraining method and selected backbone over the self-supervised and semi-supervised baselines and alternative models. Conclusion: Our findings demonstrate that foundation models, adapted and pretrained using our novel approach, can effectively classify REIMS data even with limited labeled examples. This highlights the viability of foundation models for enhancing real-time surgical margin assessment, particularly in data-scarce clinical environments.

Enhanced-accuracy ion-range verification in real time shall enable a significant step forward in the use of therapeutic ion beams. Positron-emission tomography (PET) and prompt-gamma imaging (PGI) are two of the most promising and researched methodologies, both of them with their own advantages and challenges. Thus far, both of them have been explored for ion-range verification in an independent way. However, the simultaneous combination of PET and PGI within the same imaging framework may open-up the possibility to exploit more efficiently all radiative emissions excited in the tissue by the ion beam. Here we report on the first pre-clinical implementation of an hybrid PET-PGI imaging system, hereby exploring its performance over several ion-beam species (H, He and C), energies (55 MeV to 275 MeV) and intensities (10$^7$-10$^9$ ions/spot), which are representative of clinical conditions. The measurements were carried out using the pencil-beam scanning technique at the synchrotron accelerator of the Heavy Ion Therapy centre in Heidelberg utilizing an array of four Compton cameras in a twofold front-to-front configuration. The results demonstrate that the hybrid PET-PGI technique can be well suited for relatively low energies (55-155 MeV) and beams of protons. On the other hand, for heavier beams of helium and carbon ions at higher energies (155-275 MeV), range monitoring becomes more challenging owing to large backgrounds from additional nuclear processes. The experimental results are well understood on the basis of realistic Monte Carlo (MC) calculations, which show a satisfactory agreement with the measured data. This work can guide further upgrades of the hybrid PET-PGI system towards a clinical implementation of this innovative technique.

Purpose: To identify a clinically interpretable subset of survival-relevant features in HN cancer using Bayesian Network (BN) and evaluate its prognostic and causal utility. Methods and Materials: We used the RADCURE dataset, consisting of 3,346 patients with H&N cancer treated with definitive (chemo)radiotherapy. A probabilistic BN was constructed to model dependencies among clinical, anatomical, and treatment variables. The Markov Blanket (MB) of two-year survival (SVy2) was extracted and used to train a logistic regression model. After excluding incomplete cases, a temporal split yielded a train/test (2,174/820) dataset using 2007 as the cutoff year. Model performance was assessed using area under the ROC curve (AUC), C-index, and Kaplan-Meier (KM) survival stratification. Model fit was further evaluated using a log-likelihood ratio (LLR) test. Causal inference was performed using do-calculus interventions on MB variables. Results: The MB of SVy2 included 6 clinically relevant features: ECOG performance status, T-stage, HPV status, disease site, the primary gross tumor volume (GTVp), and treatment modality. The model achieved an AUC of 0.65 and C-index of 0.78 on the test dataset, significantly stratifying patients into high- and low-risk groups (log-rank p < 0.01). Model fit was further supported by a log-likelihood ratio of 70.32 (p < 0.01). Subgroup analyses revealed strong performance in HPV-negative (AUC = 0.69, C-index = 0.76), T4 (AUC = 0.69, C-index = 0.80), and large-GTV (AUC = 0.67, C-index = 0.75) cohorts, each showing significant KM separation. Causal analysis further supported the positive survival impact of ECOG 0, HPV-positive status, and chemoradiation. Conclusions: A compact, MB-derived BN model can robustly stratify survival risk in HN cancer. The model enables explainable prognostication and supports individualized decision-making across key clinical subgroups.

Accurate assessment of tissue perfusion is crucial in visceral surgery, especially during anastomosis. Currently, subjective visual judgment is commonly employed in clinical settings. Hyperspectral imaging (HSI) offers a non-invasive, quantitative alternative. However, HSI imaging lacks continuous integration into the clinical workflow. This study presents a hyperspectral light field system for intraoperative tissue oxygen saturation (SO2) analysis and visualization. We present a correlation method for determining SO2 saturation with low computational demands. We demonstrate clinical application, with our results aligning with the perfusion boundaries determined by the surgeon. We perform and compare continuous perfusion analysis using two hyperspectral cameras (Cubert S5, Cubert X20), achieving processing times of < 170 ms and < 400 ms, respectively. We discuss camera characteristics, system parameters, and the suitability for clinical use and real-time applications.

Background: Deep learning has significantly advanced medical image analysis, with Vision Transformers (ViTs) offering a powerful alternative to convolutional models by modeling long-range dependencies through self-attention. However, ViTs are inherently data-intensive and lack domain-specific inductive biases, limiting their applicability in medical imaging. In contrast, radiomics provides interpretable, handcrafted descriptors of tissue heterogeneity but suffers from limited scalability and integration into end-to-end learning frameworks. In this work, we propose the Radiomics-Embedded Vision Transformer (RE-ViT) that combines radiomic features with data-driven visual embeddings within a ViT backbone. Purpose: To develop a hybrid RE-ViT framework that integrates radiomics and patch-wise ViT embeddings through early fusion, enhancing robustness and performance in medical image classification. Methods: Following the standard ViT pipeline, images were divided into patches. For each patch, handcrafted radiomic features were extracted and fused with linearly projected pixel embeddings. The fused representations were normalized, positionally encoded, and passed to the ViT encoder. A learnable [CLS] token aggregated patch-level information for classification. We evaluated RE-ViT on three public datasets (including BUSI, ChestXray2017, and Retinal OCT) using accuracy, macro AUC, sensitivity, and specificity. RE-ViT was benchmarked against CNN-based (VGG-16, ResNet) and hybrid (TransMed) models. Results: RE-ViT achieved state-of-the-art results: on BUSI, AUC=0.950+/-0.011; on ChestXray2017, AUC=0.989+/-0.004; on Retinal OCT, AUC=0.986+/-0.001, which outperforms other comparison models. Conclusions: The RE-ViT framework effectively integrates radiomics with ViT architectures, demonstrating improved performance and generalizability across multimodal medical image classification tasks.

Phase-contrast and dark-field imaging are relatively new X-ray imaging modalities that provide additional information to conventional attenuation-based imaging. However, this new information comes at the price of a more complex acquisition scheme and optical components. Among the different techniques available, such as Grating Interferometry or Edge Illumination, modulation-based and more generally single-mask/grid imaging techniques simplify these new procedures to obtain phase and dark-field images by shifting the experimental complexity to the numerical post-processing side. This family of techniques involves inserting a membrane into the X-ray beam that locally modulating the intensity to create a pattern on the detector which serves as a reference. However, the topological nature of the mask used seems to determine the quality of the reconstructed phase and dark-field images. We present in this article an in-depth study of the impact of the membrane parameters used in a single mask imaging approach. A spiral topology seems to be an optimum both in terms of resolution and contrast-to-noise ratio compared to random and regular patterns.

Achieving high-resolution optical imaging deep within heterogeneous and scattering media remains a fundamental challenge in biological microscopy, where conventional techniques are hindered by multiple light scattering and absorption. Here, we present a non-invasive imaging approach that harnesses the nonlinear response of luminescent labels in conjunction with the statistical and spatial properties of speckle patterns - an effect of random light interference. Using avalanching nanoparticles (ANPs) with strong photoluminescence nonlinearity, we demonstrate that random speckle illumination can be converted into a single, localized, sub-diffraction excitation spot. This spot can be scanned across the sample using the angular memory effect, enabling high-resolution imaging through a scattering layer. Our method is general, fast, and cost-effective. It requires no wavefront shaping, no feedback, and no reconstruction algorithm, offering a powerful new route to deep, high-resolution imaging through complex media.

Intensity-modulated proton therapy (IMPT) offers superior dose conformity with reduced exposure to surrounding healthy tissues compared to conventional photon therapy. Improving IMPT delivery efficiency reduces motion-related uncertainties, enhances plan robustness, and benefits breath-hold techniques by shortening treatment time. Among various factors, energy switching time plays a critical role, making energy layer optimization (ELO) essential. This work develops an energy layer optimization method based on mixed integer model and variational quantum computing algorithm to enhance the efficiency of IMPT. The energy layer optimization problem is modeled as a mixed-integer program, where continuous variables optimize the dose distribution and binary variables indicate energy layer selection. To solve it, iterative convex relaxation decouples the dose-volume constraints, followed by the alternating direction method of multipliers (ADMM) to separate mixed-variable optimization and the minimum monitor unit (MMU) constraint. The resulting beam intensity subproblem, subject to MMU, either admits a closed-form solution or is efficiently solvable via conjugate gradient. The binary subproblem is cast as a quadratic unconstrained binary optimization (QUBO) problem, solvable using variational quantum computing algorithms. With nearly the same plan quality, the proposed method noticeable reduces the number of the used energies. For example, compared to conventional IMPT, QC can reduce the number of energy layers from 61 to 35 in HN case, from 56 to 35 in lung case, and from 59 to 32 to abdomen case. The reduced number of energies also results in fewer delivery time, e.g., the delivery time is reduced from 100.6, 232.0, 185.3 seconds to 90.7, 215.4, 154.0 seconds, respectively.

Single domain generalization (SDG) has recently attracted growing attention in medical image segmentation. One promising strategy for SDG is to leverage consistent semantic shape priors across different imaging protocols, scanner vendors, and clinical sites. However, existing dictionary learning methods that encode shape priors often suffer from limited representational power with a small set of offline computed shape elements, or overfitting when the dictionary size grows. Moreover, they are not readily compatible with large foundation models such as the Segment Anything Model (SAM). In this paper, we propose a novel Mixture-of-Shape-Experts (MoSE) framework that seamlessly integrates the idea of mixture-of-experts (MoE) training into dictionary learning to efficiently capture diverse and robust shape priors. Our method conceptualizes each dictionary atom as a shape expert, which specializes in encoding distinct semantic shape information. A gating network dynamically fuses these shape experts into a robust shape map, with sparse activation guided by SAM encoding to prevent overfitting. We further provide this shape map as a prompt to SAM, utilizing the powerful generalization capability of SAM through bidirectional integration. All modules, including the shape dictionary, are trained in an end-to-end manner. Extensive experiments on multiple public datasets demonstrate its effectiveness.

Purpose: To propose a flexible and scalable imaging transformer (IT) architecture with three attention modules for multi-dimensional imaging data and apply it to MRI denoising with very low input SNR. Methods: Three independent attention modules were developed: spatial local, spatial global, and frame attentions. They capture long-range signal correlation and bring back the locality of information in images. An attention-cell-block design processes 5D tensors ([B, C, F, H, W]) for 2D, 2D+T, and 3D image data. A High Resolution (HRNet) backbone was built to hold IT blocks. Training dataset consists of 206,677 cine series and test datasets had 7,267 series. Ten input SNR levels from 0.05 to 8.0 were tested. IT models were compared to seven convolutional and transformer baselines. To test scalability, four IT models 27m to 218m parameters were trained. Two senior cardiologists reviewed IT model outputs from which the EF was measured and compared against the ground-truth. Results: IT models significantly outperformed other models over the tested SNR levels. The performance gap was most prominent at low SNR levels. The IT-218m model had the highest SSIM and PSNR, restoring good image quality and anatomical details even at SNR 0.2. Two experts agreed at this SNR or above, the IT model output gave the same clinical interpretation as the ground-truth. The model produced images that had accurate EF measurements compared to ground-truth values. Conclusions: Imaging transformer model offers strong performance, scalability, and versatility for MR denoising. It recovers image quality suitable for confident clinical reading and accurate EF measurement, even at very low input SNR of 0.2.