Monitoring stress biomarkers (dopamine, norepinephrine, adrenaline, and cortisol) and renal function indicators, like creatinine, in biological fluids, such as saliva, is crucial for non-invasive medical diagnostics. This work presents a portable, array-based sensor device designed for point-of-care (PoC) applications, combining ease of use, versatility, and on-site analysis capability. The system operates via optical output using fluorescent chemical probes—BODIPY, naphthalimides, and rhodamine—selected for their spectral properties and ability to interact with target analytes through non-covalent interactions. Probe–analyte interactions were tested in the solid state using an optical fiber to directly acquire fluorescence emission spectra. After validation in an aqueous solution, the device successfully detected dopamine, cortisol, adrenaline, and creatinine in real saliva samples. The system delivers quantitative results in under 40 minutes, with a detection limit in the pM range, and enables clear discrimination between cortisol, dopamine, and adrenaline. Thanks to its compact design and compatibility with smartphones or portable optical fibers, it allows rapid, real-time data acquisition without complex sample preparation or specialized personnel. The procedure includes 1) calibration in water; 2) calibration in artificial saliva; 3) measurements on real saliva; and 4) validation against ELISA. By relying on chemometric interpretation of fluorescence patterns from cross-reactive probes, the device avoids signal amplification, minimizing false positives. With a cost of approximately 5 EUR per test and no need for sample pre-treatment, it offers a cost-effective, scalable solution for clinical settings.

Jet classification plays a vital role in high-energy physics by helping distinguish between fundamental particles such as quarks and gluons in particle collision events. This study explores the potential of deep learning models—specifically, a custom Convolutional Neural Network (CNN) and a pre-trained ResNet50—in classifying quark and gluon jets using 2D histogram representations derived from Pythia8 simulations. The goal is to evaluate the performance of these models in a binary classification task using only visual features, without incorporating high-level physical attributes like jet substructures.

The dataset consists of 40,000 samples (20,000 quark and 20,000 gluon jets), transformed into normalized 64×64 grayscale histograms. The CNN was carefully tuned for optimal architecture and regularization, while the ResNet50 was applied with matching parameters to allow fair comparison. Performance was evaluated using precision, recall, F1-score, and ROC-AUC metrics.

Results showed that both models achieved moderate classification performance, with the CNN slightly outperforming ResNet50. The best CNN configuration reached an accuracy of 68.2% and an AUC of 0.74, surpassing random guessing but falling short of deployment-ready reliability.

This work highlights both the potential and limitations of image-based jet classification using deep learning. It also emphasizes the importance of model architecture, preprocessing, and domain-specific features. The findings serve as a stepping stone for future research on integrating physics-aware features and advanced architectures for more robust particle identification.

The study of sodium diclofenac (SDF) adsorption on alginate/inorganic filler(Alg/PZ) microcapsules addresses the environmental issues related to the presence of pharmaceutical contaminants in water. The Alg/PZ biomaterial microcapsules were synthesized by crosslinking under various conditions (polymer: 0.75-1.3 g; inorganic filler: 0.4-1.2 g). The product and raw materials were characterized by thermogravimetric analysis (TGA); Fourier transform infrared spectroscopy (FTIR); and scanning electron microscopy (SEM).

TGA revealed optimal encapsulation or synthesis for the 0.75/1.2 mass ratio (Alg/PZ), with 63.87% PZ encapsulation, indicating enhanced thermal stability. Effective entrapment of PZ within the polymeric matrix was proven by FTIR and TGA analyses, highlighting hydrogen and electrostatic bonds, while SEM images confirmed a spherical, uniform, and porous bead morphology with a diameter of 1.73 mm. Sallow-bed adsorption revealed instantaneous pseudo-second-order kinetics, reaching equilibrium in 45 minutes, with a maximum adsorption capacity (q_max) of 21 mg·g^-1 for an adsorbent mass of 0.2 g. Furthermore, kinetic studies, supported by the Weber–Morris and Crank (squared driving force model) models, highlighted pore accessibility as well as a concentration-dependent effective diffusion coefficient (Dₑff). The squared driving force mass transfer model validated diffusion-limited kinetics (D_ₑff = 2.96×10⁻⁷ cm²·s⁻¹, R² = 0.9832). This study validates Alg/PZ composites as sustainable and scalable solutions compared to conventional adsorbents, enabling 98% SDF removal under optimal conditions.

Human hemoglobin is a tetramer consisting of two α and two β subunits. Each subunit contains one identical ferrous heme group that can reversibly bind one ligand such as carbon monoxide (CO). Determining CO dynamics in hemoglobin subunits is essential for gaining insight into the transport of small molecules in physiological systems. Here, we use picosecond to millisecond transient mid-infrared (mid-IR) spectroscopy to study the photoinduced dynamics of CO in isolated hemoglobin subunits. Photoinduced absorption changes of the isolated carbonmonoxy hemoglobin subunits were measured after photoexcitation at 543 nm into the Q bands of the heme moiety. Time-resolved spectra in the mid-IR region were measured on the ULTRA apparatus at the Central Laser Facility (Didcot, UK). All the experiments were performed in 50 mM Tris buffer, pD 8.2, at 19°C.

The time evolution of the vibrational spectra of the coordinated and as photodissociated CO molecules was monitored in the spectral range between 1900 and 2180 cm^-1. The mid-IR spectrum of the liganded subunits shows discrete CO stretch bands, denoted A₀ (~1,968 cm^–1) and A₁ (~1,950 cm^–1). The distinct stretch bands for CO photolyzed and temporarily trapped in the protein matrix are detected in the region of 2,090–2,160 cm^–1. The measured transient mid-IR spectra were analyzed using singular value decomposition and maximum entropy method analysis. We succeeded in following the evolution of CO in hemoglobin subunits. The kinetic model, describing both the photodissociation and subsequent rebinding of CO, is introduced and discussed.

Predicting forest fire occurrence is essential for proactive disaster preparedness and environmental protection. We introduce a machine learning-based system that forecasts next-day fire probability at high spatial resolution using satellite-derived, multi-modal geospatial data. In contrast to existing reactive systems that rely on thermal anomaly detection (e.g., MODIS or VIIRS-SNPP), our approach is fully predictive, generating pixel-wise fire risk maps a day in advance. Our study focuses on Uttarakhand, India, which is an ecologically sensitive region that experiences frequent and severe forest fires. We curated a domain-specific geospatial dataset spanning 1 April to 29 May 2016. It includes daily 30-meter GeoTIFF images with 10 bands comprising weather (e.g., temperature, wind, precipitation), topography (slope, aspect), fuel map, and fire mask. We constructed this dataset from diverse sources and aligned all bands spatially and temporally. To demonstrate the usefulness of this dataset, we implement a deep convolutional neural network (CNN) using the ResUNet-a architecture, chosen for its robust performance in the semantic segmentation of high-resolution remote sensing data. Our model is trained from scratch to produce high-resolution fire probability maps and classify fire/no-fire pixels. Our solution helps with planning and decision-making for early intervention, especially in areas with high risk. It supports UN’s SDG 13 (Climate Action) and SDG 15 (Life on Land) by enhancing resilience and conserving ecosystems. The presented dataset and methodology can serve as a benchmark for future research on wildfire risk prediction using Earth observation data.

Synthetic azo dyes including red Ponceau 4R (E124) are widely applied in the food industry to provide a bright, attractive, and stable color to foodstuff. Nevertheless, negative health effects can appear with high dye consumption. Therefore, the Ponceau 4R content in foods is strictly controlled. Voltammetric sensors are a promising tool for solving this problem. A glassy carbon electrode with layer-by-layer coverage of multi-walled carbon nanotubes and cerium dioxide nanoparticle dispersion in cetylpyridinium bromide has been designed as a novel sensitive voltammetric sensor for Ponceau 4R. The modifier combination provides an 80 mV cathodic shift of the Ponceau 4R oxidation peak and 1.7-fold higher oxidation currents compared to the bare GCE due to the synergistic effect of nanomaterials. Moreover, the appearance of cathodic step on the Ponceau 4R voltammograms indicates an increase in the electron transfer rate that is clearly confirmed by the electrochemical impedance spectroscopy data (k_et = 1.15 × 10^–3 and 5.19 × 10^–5 cm s^–1 for the modified and bare GCE, respectively). Ponceau 4R electrooxidation is a quasi-reversible, pH-independent, and diffusion-driven process. Differential pulse mode (pulse amplitude of 75 mV and time of 25 ms) in Britton–Robinson buffer pH 2.0 has been applied for Ponceau 4R quantification. The sensor response at 0.76 V is linear in the range of 0.10–1.0 and 1.0–7.5 μM of Ponceau 4R with the detection limit of 23 nM, which is sufficient for practical application. Recovery of dye in model solutions (99-101%) confirms high accuracy of the sensor developed.

This study focuses on the chemical synthesis of high-purity silica derived from Algerian diatomite, targeting its use in advanced technological applications, particularly in the photovoltaic sector. Diatomite, a naturally abundant siliceous mineral, was subjected to a controlled chemical purification process using a 4 mol/L sodium hydroxide (NaOH) solution. This alkaline leaching aims to dissolve impurities, enhance silica content, and improve the material's structural order. The treated samples were extensively characterized by X-ray diffraction (XRD), to assess crystallinity and phase evolution, and X-ray photoelectron spectroscopy (XPS), to examine surface chemistry and impurity levels. The untreated diatomite contained approximately 80% SiO₂. Post-treatment, XRD patterns indicated a clear improvement in crystallinity and the removal of amorphous or non-siliceous phases. XPS analysis revealed a marked reduction in surface impurities, particularly metallic elements and carbon-containing species, with a significant drop in carbon peak intensity, indicating a cleaner silica surface. This enhanced surface purity is crucial for the downstream production of solar-grade silicon. Overall, the results confirm that sodium hydroxide NaOH-based chemical treatment is a simple, effective, and scalable approach for producing high-purity silica from natural diatomite. The purified material demonstrates excellent potential for use in the photovoltaic industry and other advanced applications requiring ultra-clean silica sources.

Early and accurate diagnosis of kidney cancer is important for effective treatment planning and improved patient prognosis. This work proposed a strong ensemble of deep learning for binary classification of kidney histopathological images into tumor and normal classes. The dataset employed was obtained from the publicly accessible Multi Cancer Dataset on Kaggle, with all images resized to 128×128 pixels to ensure consistency. We implemented a bagging ensemble strategy by training three distinct convolutional neural network models, each based on a pre-trained ResNet50 architecture with frozen base layers. Each model was trained on a different subset of the training data to promote diversity within the ensemble. The predictions of each model were aggregated with soft voting for the final prediction. Based on the evaluation of the test set, our ensemble achieved an accuracy of 94.46% with high precision, recall, and F1-scores. Our results demonstrate that the bagging ensemble effectively has robustness in automated kidney cancer detection and has potential as a decision-support tool in clinical practice. The proposed method not only reduces variance and improves classification stability but also highlights the effectiveness of using ensemble learning with transfer learning for histopathological image analysis, as we note that it was a successful collaboration.

The proliferation of AI-generated synthetic faces has raised significant concerns about media authenticity, identity theft, and misinformation. Detecting such fake faces reliably is critical for securing biometric systems and restoring public trust in digital content. This study investigates the effectiveness of transfer learning using the Xception model—a deep convolutional neural network originally trained on ImageNet—for binary classification of real versus AI-generated face images.

We combined and preprocessed two publicly available datasets, resulting in a balanced corpus of authentic and synthetic face images. The data was resized to 299×299 pixels, normalized, and split into training (70%), validation (20%), and test (10%) sets. A fully unfrozen Xception model was fine-tuned using an optimized architecture and trained over 30 epochs with the Adam optimizer and binary cross-entropy loss. Performance was evaluated using accuracy, precision, recall, F1-score, and confusion matrix, along with qualitative analysis through prediction visualizations.

The fine-tuned model achieved nearly 100% training accuracy and 90% validation accuracy. On the unseen test set of 1,205 images, it attained 89.88% accuracy and an F1-score of 0.90, indicating high reliability across both real and fake face classes. The model performed slightly better at identifying synthetic faces, highlighting detectable artifacts introduced during generation.

Our findings confirm that transfer learning with the Xception model is a practical, reproducible solution for fake face detection, even in resource-limited academic settings. This study contributes a streamlined pipeline and benchmarks for future work in visual deepfake detection and media forensics.

This study proposes an innovative and ecological approach for the synthesis of carbon-contained composite foams from glucose through thermal decomposition under pressure. Calcium chloride reaction through foam treatment in CO₂- NH₃-H₂O vapors provides additional functions of the material. CaCl₂ conversion enables advanced gas absorption applications. These findings clearly demonstrate the material’s capacity for effective absorption of both water vapor and volatile acids concurrently allowing their use as phase transformation materials (PCMs). The porous and hydrophilic nature of the carbon foam increases the contact surface of the concentrated CaCl₂ solution-wet air interface, increasing the absorption speed of water vapors and simultaneously preventing the leakage of the formed liquid. The compression testing shows that the mechanical strength of the foam is influenced by the glucose – CaCl₂ ratio. The X-ray diffraction and Fourier Transform Infrared analyses confirmed the compositional changes in the foam after CO₂ treatment. According to X-ray diffraction analysis, the formation of calcite as a single crystalline phase, can be clearly observed. After prolonged heating of the foams initially treated in CO₂ and NH₃ vapors, no traces of NH₄Cl are detected, indicating its complete sublimation during the thermal treatment. The results obtained show that the carbon foams synthesized by this method have multiple uses, including different gas capture and thermal management. This work was supported by a grant of the Ministry of Research, Innovation, and Digitization, CCCDI-UEFISCDI, project number PN-IV-P8-8.3-PM-RO-BE-2024-0004 within PNCDI IV.