Atmospheric bioaerosol inactivation is a critical research priority for controlling the transmission of airborne pathogens. However, conventional disinfection systems often exhibit slow bactericidal kinetics and lack quantitative frameworks to characterize the precise dose–response relationship. To address these limitations, we proposed a decoupled composite disinfection system (DCD-System) integrating non-thermal plasma (NTP), ultraviolet (UV) radiation, and photocatalysis. This study provided two primary contributions: 1) the introduction of a time-integrated dose–survival model to accurately quantify inactivation kinetics, and 2) the evaluation of an integrated oxidation framework that accelerated irreversible ultrastructural damage. The plasma module of the DCD-System featured a localized intense electric field of 2.2 × 107 V/m and an electron density of 1.9 × 1012 cm-3, and generated reactive oxygen species (ROS, including O2- and 1O2). Evaluations using Staphylococcus albus in a 20 m3 chamber demonstrated that the DCD-System achieved a log reduction > 4.0 within a 15 min effective dose period, significantly outperforming standalone modules. Furthermore, the DCD-System exhibited a lower median inhibitory dose (ID50), indicating that combined exposure to plasma-generated ROS and UV radiation accelerated cell membrane permeability and intracellular protein leakage.
Computed tomography (CT) plays a vital role in clinical diagnosis, yet reducing radiation exposure while maintaining image quality remains a significant challenge. Although data-driven methods have demonstrated effectiveness, their practical application is often hindered by the reliance on large paired datasets, which are difficult to obtain in clinical practice. This paper introduces a data distribution consistency constraint, inspired by the entropic regularization of optimal transport (OT), into the Kantorovich problem. To this end, an unsupervised framework is established for low-dose CT reconstruction, in which the data consistency conditions are integrated into both the image domain and projection domain. To validate the proposed model, we design a network that combines generative adversarial networks with OT theory, incorporating dedicated modules for projection-domain detail restoration, image-domain noise reduction, and image reconstruction. In simulation experiments with the lowest photon dose (1×10³), the proposed method achieves a 0.5dB improvement in peak signal-to-noise ratio compared to baseline algorithms and significantly enhances structural similarity, increasing from 0.81 to 0.88. Real-data experiments under three X-ray tube current settings further demonstrate the method’s promising noise reduction and detail recovery capabilities. Both theoretical analysis and experimental results suggest that the proposed framework significantly enhances noise reduction and detail restoration in LDCT reconstructed images without requiring paired datasets.
Objective: This study investigates the therapeutic potential of cold atmospheric plasma (CAP) in promoting chronic wound healing in diabetic mice and explores the associated cellular responses. Methods: In vitro experiments were conducted using HaCaT and RAW 264.7 cells to evaluate the effects of CAP on cell viability, migration, and proliferation. Transcriptomic analysis was performed in RAW 264.7 cells to identify CAP-induced gene expression changes. In vivo, a streptozotocin-induced diabetic mouse model with full-thickness skin wounds was treated with CAP for 0 s- 40 s, and wound healing was assessed through histological and immunohistochemical analyses. Results: CAP treatment significantly enhanced cell viability, migration, and proliferation in vitro. RNA sequencing revealed enrichment of genes associated with DNA replication and cell cycle regulation. In diabetic mice, CAP treatment accelerated wound closure and promoted re-epithelialization, collagen deposition, and cellular proliferation at the wound site. Conclusion: CAP treatment promotes diabetic wound healing by enhancing cellular proliferation and migration while modulating local inflammatory responses. Significance: These findings support CAP as a promising non-invasive therapeutic strategy for chronic diabetic wound management.
Image reconstruction in positron emission tomography (PET) is essential for disease diagnosis and treatment evaluation, but it struggles with noise and inadequate image quality. The traditional kernel expectation maximization (KEM) algorithm uses prior information to improve reconstruction quality and offers computational simplicity. However, its fixed kernel forms, noise sensitivity, and limited flexibility might adversely affect the accuracy and robustness of the reconstructed images. In this study, we propose a bi-level optimization model that alternately refines a multi-kernel matrix and a diffusion coefficient prior to capture non-linear mappings from projection data. Specifically, we employ a pre-trained score-based diffusion model to achieve improved noise suppression and detail recovery. Guided by reverse stochastic differential equations, this model effectively captures the complex distribution. Additionally, a multi-kernel strategy combining radial Gaussian and polynomial kernels is utilized to enhance the optimization of the kernel matrix. Experimental results across multiple datasets demonstrated that the pro-posed model outperformed existing kernel-related methods in dynamic PET reconstruction. Preliminary clinical experiments under a one-half-count, slice-wise 2D setting demonstrated the feasibility of the proposed framework, while more severe count-reduction and fully 3D evaluations remain necessary.
X-ray CT often suffers from shadowing and streaking artifacts in the presence of metallic materials, which severely degrade imaging quality. Physically, the linear attenuation coefficients (LACs) of metals vary significantly with X-ray energy, causing a nonlinear beam hardening effect (BHE) in CT measurements. Reconstructing CT images from metal-corrupted measurements consequently becomes a challenging nonlinear inverse problem. Existing high-performance metal artifact reduction (MAR) methods are based on supervised learning and often require paired training data, while many unsupervised methods typically still rely on large external artifact-free datasets and generally achieve lower performance than supervised techniques. In this work, we propose Density neural representation (Diner), a novel unsupervised MAR method. Our key innovation lies in formulating MAR as an energy-independent density reconstruction problem that strictly adheres to the photon-tissue absorption physical model. This model is inherently nonlinear and complex, making it a rarely considered approach in inverse imaging problems. By introducing the watere-quivalent tissues approximation and a new polychromatic model to characterize the nonlinear CT acquisition process, we directly learn the neural representation of the density map from raw measurements without using external training data. This energy-independent density reconstruction framework fundamentally resolves the nonlinear BHE, enabling superior MAR performance across a wide range of scanning scenarios. Extensive experiments on both simulated and real-world datasets demonstrate the superiority of our unsupervised Diner over popular supervised methods in terms of MAR performance and robustness. To the best of our knowledge, Diner is the first unsupervised MAR method to outperform its supervised counterparts.
Semi-supervised learning (SSL) has shown notable potential in relieving the heavy demand of dense prediction tasks on large-scale well-annotated datasets, especially for the challenging multi-organ segmentation (MoS). However, the prevailing class-imbalance problem in MoS caused by the substantial variations in organ size exacerbates the learning difficulty of the SSL network. To address this issue, in this paper, we propose an innovative semi-supervised network with BAlanced Subclass regularIzation and semantic-Conflict penalty mechanism (BASIC) to effectively learn the unbiased knowledge for semi-supervised MoS. Concretely, we construct a novel auxiliary subclass segmentation (SCS) task based on priorly generated balanced subclasses, thus deeply excavating the unbiased information for the main MoS task with the fashion of multi-task learning. Additionally, based on a mean teacher framework, we elaborately design a balanced subclass regularization to utilize the teacher predictions of SCS task to supervise the student predictions of MoS task, thus effectively transferring unbiased knowledge to the MoS subnetwork and alleviating the influence of the class-imbalance problem. Considering the similar semantic information inside the subclasses and their corresponding original classes (i.e., parent classes), we devise a semantic-conflict penalty mechanism to give heavier punishments to the conflicting SCS predictions with wrong parent classes and provide a more accurate constraint to the MoS predictions. Extensive experiments conducted on two publicly available abdominal datasets, i.e., the WORD dataset and the MICCAI FLARE 2022 dataset, have verified the superior performance of our proposed BASIC compared to other state-of-the-art methods.
Text-guided medical image segmentation methods have significant potential to reduce the reliance on labeled data. However, existing methods often struggle to balance the accurate depiction of local details (such as lesion boundaries) with the semantic consistency of global structures (such as the lesion count and location as described in medical reports). To address this issue, we propose a Dual-Granularity Text-Guided Network for medical image segmentation, named DGTG-Net, which synergistically leverages textual prior knowledge at different granularities through a unified architecture, guiding the segmentation process from both local and global perspectives. First, we construct a local micro-pathway during encoding, which employs the Semantic Information Injection Module with an Adaptive Gating Mechanism to dynamically filter textual noise and accurately enhance fine-grained visual features using word-level tokens. Second, in the decoding stage, we establish a global macro-pathway that continuously injects sentence-level embeddings via a Text-Guided Decoder. This ensures strong macro-semantic constraints and provides structural correction throughout mask reconstruction. Extensive experiments on three diverse datasets (QaTa-COV19, MosMedData+, and Kvasir-SEG) show that DGTG-Net achieves the best performance on QaTa-COV19, MosMedData+ and the second-best performance on Kvasir-SEG. Furthermore, when trained with only 75% of the labeled data, DGTG-Net outperforms nnU-Net trained with 100% of the annotations across all three datasets, suggesting improved label efficiency relative to this vision-only baseline.
Atmospheric-pressure plasma jet (APPJ) have emerged as a transformative nonthermal modality, enabling controlled delivery of reactive species, electric fields, and ultraviolet radiation for targeted biomedical intervention. Floating-electrode tube APPJ (FTAPPJ) extend this capability to remote plasma delivery with strong translational potential, yet their development is constrained by insufficient understanding of discharge regulation, species transport, and engineering reliability. Here, an FTAPPJ was systematically investigated using a two-dimensional fluid model, optical and electrical diagnostics, gas- and liquid-phase chemical measurements, and U87-MG glioma cell assays. The applied voltage governed floating-electrode charging through a self-limiting saturation mechanism, whereas tip curvature affected only the local near field. More importantly, an inner–outer decoupled reaction-regulation mode was identified, in which ionization and precursor formation occurred primarily inside the tube, while reactions with ambient gases dominated downstream. Accordingly, O2 admixture reconfigured rather than uniformly enhanced reactive-species delivery, suppressing electron density and He∗ while increasing atomic O and O3 and decreasing liquid-phase NO−2 and H2O2. Among the tested conditions, 0.5% O2 produced the strongest inhibition of U87-MG cell viability, indicating an optimal balance between discharge sustainment and biologically effective oxidant delivery. These findings provide a rational design basis for clinically translatable FTAPPJ platforms for remote biomedical applications.
The interest in Compton cameras in medical imaging keeps rising given their potential advantages over gamma cameras in the case of low activities or high photon energies. The IRIS group of IFIC (Valencia, Spain) has developed a Compton camera prototype for medical imaging, currently being tested for nuclear medicine applications. Initial tests carried out by the group with different medical radiotracers in phantoms and patients pointed out the necessity of a system with improved detection efficiency. For this purpose, MACACO III+, a new prototype version with an enlarged second detector area, has been assembled. Characterization tests in the laboratory confirm the correct system operation and the increase in sensitivity. Experimental tests with the system have been conducted at La Fe Hospital (Valencia, Spain) with 131I-MIBG using Derenzo-like and for the first time, irregular, thyroid-shaped phantoms. Rods of 4 mm diameter have been resolved experimentally at 364 keV, and the results are also accurately reproduced by GATE simulations. Thyroid-shaped phantom tests showed that MACACO III+ can reproduce organ shapes containing uniform and heterogeneous activity distributions.
Dual-ended readout monolithic PET detectors provide high detection efficiency, DOI capability, and reduced edge-related positioning degradation, but require readout electronics capable of channel compression, dual-ended event association, and scalable event processing. This work presents a hierarchical FPGA-based electronics architecture with discrete analog frontend readout for a dual-ended monolithic PET detector. The system consists of signal acquisition, module-level preprocessing, and centralized processing layers. Row-column summation multiplexing is used to reduce the SiPM energy-readout channels, while FPGA-based processing performs energy extraction, adaptive-threshold position calculation, top/bottom event matching, dualended event fusion, and full-ring coincidence processing. Experimental results show that the proposed readout scheme reduces the effective energy-channel count by 83.3%. The detector achieves an energy resolution of 12.20% FWHM at 511 keV, a transverse spatial resolution of 0.97 mm, and a DOI resolution of 1.39 mm. The preprocessing stage provides a valid dual-ended event output rate of approximately 1 Mcps. An electrical split-delay benchmark was also performed to verify the FPGA-based timestamp extraction path; this benchmark is not interpreted as the timing resolution of the complete detector module. These results demonstrate that the proposed architecture provides a feasible and scalable readout solution for dual-ended monolithic PET detector development.
Intracranial metastatic disease (IMD) from diverse extracranial primaries can exhibit 18F-fluoropivalate (FPIA)-detectable short-chain fatty acid transcellular flux, although substantial biological heterogeneity is expected across tumour types and lesions. However, regional metabolic variations within individual brain metastases have not been well characterized, especially given the challenges of obtaining spatially distinct biopsies in vivo. In this study, we applied an unsupervised time-series clustering approach to dynamic 18F-fluoropivalate positron emission tomography data to interrogate intra-tumoural short-chain fatty acid transcellular flux heterogeneity. Three distinct metabolic subpopulations (here termed imaging-defined oligoclones - IDOs) were identified within treatment-naïve brain metastases (from lung, breast, melanoma, and colorectal primaries), each defined by a unique fluoropivalate uptake and retention kinetic profile. These clusters exhibited non-overlapping spatial distributions and showed distinct imaging phenotypes, as evidenced by divergent associations with overall survival and with the fluoropivalate net influx rate constant Ki. Notably, the metabolic IDOs were intermingled throughout lesions rather than confined to tumour core or periphery. In patients who underwent stereotactic radiosurgery and were rescanned 4–8 weeks later, we observed a remodeling of the metabolic clusters, including the loss of the rapid-uptake cluster post-treatment and a predominance of the slow-retention cluster, coincident with a uniform decrease in cellularity across all clusters. Our spatiotemporal clustering analysis identified intra-tumoural metabolic subpopulations in brain metastases with distinct kinetic behaviours and exploratory associations with outcome. These findings introduce a non-invasive imaging-based framework for characterizing metabolic heterogeneity in brain metastases and suggest that such heterogeneity may be relevant to treatment response and clinical outcome.
Semi-monolithic scintillation detectors have recently proven to be a promising compromise between pixelated and monolithic crystals. This geometry is particularly suitable for preclinical Positron Emission Tomography (PET) systems, as it provides high spatial resolution while enabling Depth of Interaction (DOI) capabilities. While prior detector developments have almost exclusively focused on lutetium-based scintillators, BGO has re-gained attention for PET applications given its lower cost, higher stopping power, larger photo-fraction and absence of intrinsic radioactivity. In this work, we propose a semi-monolithic detector composed of 44 BGO slabs of 1 mm × 24.2 mm × 10 mm each, designed for preclinical PET systems. We focus on evaluating its spatial and energy performance and directly compare it with a LYSO semi-monolithic block of identical dimensions. The x-monolithic direction and DOI resolutions were evaluated using neural networks, while the y-pixelated direction was evaluated with an analytical method. Average FWHM spatial resolutions of 1.6 ± 0.2 mm and 1.2 ± 0.3 mm were achieved for the BGO and LYSO, respectively, along the x-monolithic direction. Regarding DOI, average FWHM resolution values of 3.0 ± 0.6 mm and 2.0 ± 0.5 mm were obtained for BGO and LYSO, respectively. A FWHM spatial resolution of around 1 mm was estimated for both crystal types along the y-pixelated direction. Finally, mean energy resolutions of 23.7% and 18.5% were found for the BGO and LYSO, respectively. These results show that the proposed BGO block is a viable alternative for high spatial resolution, high sensitivity and low-cost preclinical PET scanners.
Boron Neutron Capture Therapy (BNCT) is a targeted radiotherapy modality that couples neutron irradiation with selective drug delivery. After administration of a 10Benriched compound that preferentially accumulates in tumour tissue, the patient is exposed to thermal neutrons, triggering the 10B(n,α)7Li reaction. The resulting high-LET charged particles release their energy within micrometer-scale ranges, concentrating the absorbed dose at the cellular level while sparing surrounding healthy tissue. In the majority of capture events, the resulting 7Li nucleus de-excites by emitting a 478 keV prompt γ ray, which can be detected externally and used for dose localization and real-time monitoring with single-photon emission computed tomography (SPECT). In this work, we present the results of an experimental tomography study conducted with a BNCT-SPECT prototype system at the LENA neutron facility in Pavia, Italy. In this study, we introduce a new collimator and a new 3D gamma-ray position estimator. The system employs the BeNEdiCTE detection module, based on a 5 cm × 5 cm × 2 cm LaBr3(Ce+Sr) crystal coupled to an 8×8 matrix of Near Ultraviolet High-Density silicon photomultipliers (SiPMs). A redesigned lead pinhole collimator with higher geometric efficiency improves peak identification at 478 keV, reduces the module dimensions, and enables more robust acquisitions without advanced data processing. A compact, hardware-oriented artificial neural network (ANN) is introduced to simultaneously estimate the interaction coordinates (x, y, z) in the monolithic scintillator. The ANN is trained on a full experimental dataset and preserves good x and y performance around 3mm FWHM, while providing depth-of-interaction (DOI) estimation with 5mm FWHM resolution, enabling parallax-error compensation for large acceptance-angle geometries. We demonstrate a successful 3D reconstruction of two vials filled with boric acid (7371 ppm of 10B) placed 1.4 cm apart using four partial projections, confirming a system spatial resolution under 1 cm.
Glioblastoma is an aggressive, malignant cancer of the brain which creates a particularly challenging environment for non-ionizing and externally applied field therapies. This is in part because of the intracranial location, heterogeneous tumor environment and the sensitivity of the surrounding tissues. Plasma discharge tubes have been demonstrated to be a novel technology to generate broadband electromagnetic fields without plasma-generated reactive chemical species, with emissions spanning frequency regimes governed by different physical mechanisms. This work seeks to develop a patient-specific dosimetry framework for plasma electromagnetic field therapy based on a hybrid model of electroquasistatic (EQS) and radiofrequency (RF) field propagation. Contrast-enhanced magnetic resonance imaging (MRI) scans of patient’s head anatomy and target volumes were derived, and tissue properties were derived using Cole-Cole formulations from kilohertz to gigahertz frequencies. Broadband plasma emissions were spectrally weighted on a perfrequency basis, solved using the appropriate field solver, and combined into a dose term called the Field-Intensity Exposure Index; this was defined by the spatially resolved, spectrally weighted squared electric field magnitude. The dose distribution was computed within tumor and normal brain regions and summarized using dose–volume histograms (DVHs). Simulations were performed on a representative glioblastoma case, demonstrating that low frequency EQS components have deeper intracranial exposure, while the higher frequency components represent a larger magnitude of the dose. Single-source configurations produced higher mean exposure within the tumor relative to normal brain and multi-source configurations improved target coverage and homogeneity with diminishing returns beyond 4 sources. This framework provides a physically consistent definition for comparing source configurations, patient optimization and supports treatment planning and optimization for broadband plasma electromagnetic therapies.
Accurate voxel-wise personalized dosimetry is essential for maximizing therapeutic efficacy while minimizing off-target toxicity in 177Lu-PSMA radiopharmaceutical therapy. While Monte Carlo (MC) simulations can be used for dose calculations with gold-standard accuracy, their high computational cost limits clinical applicability. Conversely, analytical methods based on the Medical Internal Radiation Dose (MIRD) formalism can generate rapid dose estimates but lack spatial precision. We present DiffuDose, a novel AI-driven dosimetry framework that combines diffusion probabilistic models with multiscale information fusion and a voxel-wise fidelity constraint to generate accurate dose rate maps from post-therapy SPECT and CT images. DiffuDose is trained in two phases: a pretraining phase and an end-to-end joint training phase. Ablation studies confirm that multiscale information fusion and joint training are essential components of the framework. Evaluated on a clinical 177Lu-PSMA cohort, DiffuDose significantly outperformed existing deep learning models and MIRD calculations across multiple quantitative metrics, achieving accuracy levels comparable to MC dosimetry at a fraction of the computational cost.
High-resolution insert detectors used with positron emission tomography (PET) scanners can locally improve the spatial resolution of the system. PET detector performance depends on multiple factors, including the reflector material in the scintillator crystal arrays and the microcell size of the silicon photomul-tipliers (SiPMs). In this study, we compared dual-ended readout detectors comprising 24 × 24 LYSO arrays with 1.0-mm pitch and different reflectors—BaSO4 or Toray E60—read out by onsemi SiPMs with 35-μm microcells. We also evaluated detectors with 8 × 8 SiPM arrays from KETEK with 3.36-mm pitch and different microcell sizes—15, 25, or 50 μm—coupled to the LYSO array with BaSO4 reflector. Results showed that the BaSO4 reflector led to better detector performance than Toray, with flood quality of 2.55 ± 0.26, energy resolution of 11.9% (11.4%–13.1%), which im-proved to 9.6% (9.2% – 10.3%) after DOI-based calibration, depth-of-interaction (DOI) resolution of 2.04 ± 0.19 mm, and coin-cidence timing resolution (CTR) of 759 ± 109 ps. Among SiPMs with different microcell sizes, those with 25-μm microcells showed the best flood quality of 2.25 ± 0.32, with energy resolution of 12.7% (12.2%–13.7%) before and 10.6% (10.1% – 11.6%) after DOI-based calibration, DOI resolution of 2.03 ± 0.13 mm, and CTR of 855 ± 118 ps. Based on these results, an LYSO array with BaSO4 reflector read out by onsemi SiPM arrays with 35-μm microcells was selected as the candidate design for the high-resolution insert detectors being developed for the NeuroEXPLORER brain PET scanner.
Photon counting computing tomography (PCCT) is widely expected to replace traditional, energy integrating CT as a diagnostic imaging modality due to its many advantages. Amongst the most well known advantages is the ability to minimise/correct imaging artifacts caused by electronic noise and photon starvation, such as beam hardening and metal streak artifacts. Instead PCCT is known to suffer spectral distortions due to pulse pileup (PP) and charge sharing effects (CSE). A range of pulse processing schemes have been proposed to mitigate the impact of PP (Alternative Count Triggering Schemes, ACTS) and CSE (Charge Sharing Correction Algorithms, CSCA) on image quality. Most studies on CSCAs and ACTS focus on detector performance parameters, with less exploration of imaging implications. No studies have compared ACTS and CSCAs for their performance directly in the imaging domain. To address this, we have performed an extensive in silico trial comparing a wide range of pulse processing schemes for their impact on image quality. In total 18 different processing schemes were simulated (5 ACTs, 12 CSCAs, and the standard approach, STD), each operating on arrays of pixels spanning 100 to 400 μm in size (in 50 μm intervals). These systems were all tested at 3 increasing, medically relevant x-ray fluxes. In this work, we identify and explain novel imaging artifacts observed to arise from the counting in PCCT. We address how the various pulse processing schemes compare in their susceptibility to generating these artifacts, and the impact of them on contrast to noise ratio and contrast linearity.
Chest X-ray (CXR) analysis models deployed in clinical environments must continuously adapt to newly defined or emerging disease categories, while historical patient images are often unavailable due to privacy, storage, and regulatory constraints. This poses a practical challenge for class-incremental learning (CIL): models should learn new classes without replaying old data while preserving previously acquired diagnostic knowledge. Existing replay-based CIL methods are difficult to apply in privacy-sensitive medical scenarios, whereas vision-only replay-free methods lack stable semantic anchors and are prone to representation drift and catastrophic forgetting. To address this challenge, we propose language-guided knowledge-enhanced class-incremental learning (LK-CIL), a replay-free vision-language framework for medical image classification. LK-CIL freezes a pretrained vision transformer and introduces lightweight task adapters for parameter-efficient incremental updates. To provide stable semantic guidance, it constructs class textual prototypes from clinically reviewed LLM-generated radiographic descriptions. These prototypes guide knowledge-aware adapter merging across tasks, aligning evolving visual representations with class-level linguistic priors. During inference, LK-CIL refines class prototypes using a graph convolutional network on a class-relation graph, selects the most suitable adapter based on feature-prototype similarity, and applies a self-refined prediction strategy to improve robustness. Experiments on ChestX-ray14 show that LK-CIL achieves a favorable balance between recognition accuracy and forgetting suppression compared with state-of-the-art (SOTA) CIL methods under multiple incremental and distribution settings. Additional evaluations further support its effectiveness on an emerging COVID-19 class, multi-label CXR classification, external CXR datasets, and other medical imaging modalities, including colon histopathology and dermoscopy. These results suggest that LK-CIL provides a practical replay-free solution for continuously updating medical image classification models without storing historical patient data. Code is available at: https://github.com/Lingling-Yuan/LK-CIL.
