Kaunas University of Technology (abbreviated as KTU, Lithuanian: Kauno technologijos universitetas) is a public research university located in Kaunas, Lithuania. Established in 1922, KTU has been one of the top centers of Lithuanian science education. According to Lithuanian National University Rankings conducted in 2021, KTU was the second best university in Lithuania. The primary language of education is Lithuanian, though there are courses that are taught jointly in Lithuanian and English or solely English.
Introduction This study evaluates polymer gel dosimetry as a potential tool for three-dimensional (3D) verification of VMAT dose delivery under respiratory-like motion. Methods A VMAT plan for a lung cancer case (18 Gy, two arcs, 10 MV FFF) was delivered on a Varian TrueBeam linear accelerator under static and sinusoidal motion conditions using an ArcCHECK phantom. Normoxic polymer gels (nMAG, nPAG, and NIPAM) were investigated. One container per gel formulation was positioned at the center of the ArcCHECK phantom and irradiated under both static and motion conditions. Irradiated gel containers were read out using CT and 3T MRI to reconstruct volumetric dose maps. Dose profiles and gamma analysis were used for evaluation. Gel dosimetry results were compared with ArcCHECK–3DVH-based dose measurements. Results Polymer gel measurements (with both CT and MRI readout) showed a similar trend to ArcCHECK–3DVH results, demonstrating reduced dose accuracy under motion compared with static irradiation conditions. Motion-related effects were observed in the reconstructed dose distributions and dose profiles. Among the investigated gels, nMAG—particularly with MRI readout—showed the clearest visualization of motion-related effects under the evaluated experimental conditions with a dynamic gamma passing rate of 54.54% for MRI readout compared with 32.27% for CT readout. CT readout, especially for nPAG and NIPAM, was less sensitive to motion-related dosimetric changes. Conclusions Polymer gel dosimetry shows potential for comprehensive 3D dose verification in motion-influenced VMAT delivery. The findings support its use as a complementary approach to ArcCHECK–3DVH, with nMAG (especially with MRI readout) showing the most promising performance in this study.
Portfolio selection is a critical issue in financial management under uncertainty. In this paper, we propose a complex approach for portfolio selection with quantile return approximation. In particular, we propose a practical workflow that combines principal component analysis with quantile regression into the Quantile Regression-Principal Component Analysis (QR-PCA) framework. The use of quantile regression allows to capture asymmetric and heterogeneous conditional behavior of return distributions. This strengthens the dimensionality reduction and robust regression techniques. For comparison, we consider parametric and nonparametric approximation techniques. We also design a new performance measure called quantile ratio (qR) based on approximate quantile expectations of returns incorporated in a portfolio optimization task. The proposed model is applied to a real-world dataset of financial assets, demonstrating its effectiveness in constructing portfolios that outperform traditional portfolio models. The empirical results reveal better risk-adjusted performance compared to those optimized using traditional models.
Parkinson’s disease (PD) is a degenerative, chronic neurological condition that impairs a person’s ability to move normally. People may experience difficulties with speaking, writing, walking, or performing basic tasks if dopamine-generating neurons in the brain are injured or die. Using traditional techniques for PD analysis is time-consuming and challenging, as the evaluation process is prone to high misclassification rates. Therefore, we proposed a deep learning-based architecture for classifying PD using WiFi signals in this work. The data is generated at the initial stage using WiFi signals. After that, we proposed two deep learning architectures from scratch. The first architecture, named E3-ST transformer, is based on a three-stage encoding scheme, and the second two residual-attention-block-based networks are named PD-RAN2. Both models are trained on generated WiFi signal data, and the hyperparameters are optimized using Bayesian Optimization (BO). In the next phase, trained models are used, and deep features are incorporated, employing a new method termed serial-based attention-weighted. The fused features are finally classified using neural network classifiers. The output is in label classes such as slow walking, fast walking, sitting on a chair, standing still, and FOG episodes. The Medium Neural Network (MN2) classifier achieved the best accuracy of 97.78
A strong Community of Practice (CoP) can be powerful in supporting people to share, generate, and disseminate knowledge. This study evaluates the use of the Communities of Practice (CoP) approach for effective knowledge consolidation in the field of citizen science. Our paper offers an analysis of four CoPs that were set up as part of the European-based 3-year WeObserve project, with distinct themes of (1) co-design citizen engagement; (2) impact and value for governance; (3) interoperability and standards; and (4) the United Nations Sustainable Development Goals. Participation across the four CoPs fluctuated during their three-year life-time. Three key outcomes emerged from the CoPs. First, a joint identity and understanding were created within and across CoPs through the creation of an inception report by each CoP and through the creation of Citizen observatory (CO) vocabulary, which also served to differentiate such observatories from citizen science (CS) initiatives. Next, scientific papers and technical reports were cooperatively produced by CoP members that represent a synthesis of CoP members’ knowledge. Essential ingredients to the success of these CoPs also included extensive stakeholder engagement and the CoPs being steered by the underpinning values of the CS community. The impacts of the WeObserve CoPs range from the uptake of jointly produced publications, novel cooperative CS projects, new CoPs, joint grant proposals, and the integration of citizen science data into SDG monitoring. This evaluation highlights the diverse and transformative potential of CoPs for citizen science practice.
Photoplethysmography (PPG) is a widely used non-invasive physiological sensing technique, suitable for various clinical applications. Such clinical applications are increasingly supported by machine learning methods, raising the question of the most appropriate input representation and model choice. Comprehensive comparisons, in particular across different input representations, are scarce. We address this gap in the research landscape by a comprehensive benchmarking study covering three kinds of input representations, interpretable features, image representations and raw waveforms, across prototypical regression and classification use cases: blood pressure and atrial fibrillation prediction. In both cases, the best results are achieved by deep neural networks operating on raw time series as input representations. Within this model class, best results are achieved by modern convolutional neural networks (CNNs). but depending on the task setup, shallow CNNs are often also very competitive. We envision that these results will be insightful for researchers to guide their choice on machine learning tasks for PPG data, even beyond the use cases presented in this work.
