University of Donja Gorica

Background Dialysis is a very complex treatment that is received by around 3 million people annually. Around 10% of the death cases in the presence of the dialysis machine were due to the technical errors of dialysis devices. One of the ways to maintain dialysis devices is by using machine learning and predictive maintenance in order to reduce the risk of patient's death, costs of repairs and provide a higher quality treatment. Objective Prediction of dialysis machine performance status and errors using regression models. Method The methodology includes seven steps: data collection, processing, model selection, training, evaluation, fine-tuning, and prediction. After preprocessing 1034 measurements, twelve machine learning models were trained to predict dialysis machine performance, and temperature and conductivity error values. Results Each model was trained 100 times on different splits of the dataset (80% training, 10% testing, 10% evaluation). Logistic regression achieved the highest accuracy in predicting dialysis machine performance. For temperature predictions, Lasso regression had the lowest MSE on training data (0.0058), while Linear regression showed the highest R² (0.59). For conductivity predictions, Lasso regression provided the lowest MSE (0.134), with Decision tree achieving the highest R² (0.2036). SVM attained the lowest MSE on testing dataset, with 0.0055 for temperature and 0.1369 for conductivity. Conclusion The results of this study demonstrate that clinical engineering (CE) and health technology management (HTM) departments in healthcare institutions can benefit from proposed automated systems for advanced management of dialysis machines.

Background In May 2018, an outbreak of NDM-5-carrying Escherichia coli (NDM-5-EC) was detected at the hemato-oncology department of a tertiary care center in Austria. This report details the outbreak investigation, control measures and the whole genome sequencing (WGS) data of the outbreak isolates. Methods A total of 15 isolates (seven clinical isolates from allogenic stem cell transplant (SCT) recipients and eight wastewater isolates recovered from patients’ toilets) were analyzed by whole genome sequencing. Results Genome based typing identified two clusters of the high risk clones ST167/CT12607 and ST617/CT2791. Long-read sequencing of selected isolates from both clusters identified two different plasmids, however with a highly similar genetic context of the blaNDM-5 containing region. Genomic analysis revealed the presence of additional resistance genes, including blaCTX-M-15, and blaOXA-1, and virulence factors. Four patients were colonized with NDM-5-EC, two patients suffered bacteremia caused by the outbreak strain and two deaths were associated with an NDM-5-EC infection. The outbreak source was traced to toilet sewage pipes, which remained persistently contaminated despite extensive cleaning and disinfection. Successful eradication of NDM-5-EC from the installations required disassembly, hot water pressure washing of the sewage pipes and complete replacement of all movable parts. Additionally, colonized patients were instructed to use wheeled commodes instead of toilets, and a pre-admission screening strategy was implemented for all patients undergoing hematologic stem cell transplantation. The outbreak was successfully terminated in November 2020. Conclusion NDM-5-EC, especially high-risk clones such as ST167 and ST617, can persist in hospital wastewater systems despite cleaning and disinfection efforts and can cause prolonged outbreaks. Therefore, a comprehensive bundle of interventions like the ones applied in our study is essential, especially in clinical settings with heavily immunosuppressed patients.

This research explores the role of synthetic data in enhancing the accuracy of deep learning models for automated poultry farm management. A hybrid dataset was created by combining real images of chickens with 400 FLUX.1 [dev] generated synthetic images, aiming to reduce reliance on extensive manual data collection. The YOLOv9 model was trained on various dataset compositions to assess the impact of synthetic data on detection performance. Additionally, automated annotation techniques utilizing Grounding DINO and SAM2 streamlined dataset labeling, significantly reducing manual effort. Experimental results demonstrate that models trained on a balanced combination of real and synthetic images performed comparably to those trained on larger, augmented datasets, confirming the effectiveness of synthetic data in improving model generalization. The best-performing model trained on 300 real and 100 synthetic images achieved mAP = 0.829, while models trained on 100 real and 300 synthetic images reached mAP = 0.820, highlighting the potential of generative AI to bridge data scarcity gaps in precision poultry farming. This study demonstrates that synthetic data can enhance AI-driven poultry monitoring and reduce the importance of collecting real data.

This paper presents the integration of optical character recognition (OCR) and advanced natural language processing (NLP) models for automated handling of matrices derived from images and textual inputs, all combined within an implemented chatbot. The motivation for choosing this topic arises from the practical experiences of the authors gained while working with groups of students who encounter the concept of matrices as part of their academic responsibilities. Through the analysis of their results and classroom interactions, it was observed that many students struggle with this area. This paper presents an innovative approach to enhancing matrix problem-solving by leveraging intelligent tutoring systems supported by High-Performance Computing, aiming to improve learning efficiency and student outcomes. By combining the EasyOCR framework and the Qwen2-Math-7B-Instruct model, operations such as transposition, addition, and multiplication of matrices are enabled. The system supports the input of one or two matrices, allowing the selection of operations through textual or image-based queries. The OCR component extracts numerical data from images, while the NLP model interprets user requests and executes operations accurately. The interface allows the addition of a second matrix image only when necessary, enhancing the system's intuitiveness and efficiency. The results of the recognition accuracy of the OCR model of image input matrices of different dimensions show a high level of accuracy of 95%, while for 2x2 matrices they reach an accuracy of 99%. This work contributes to the development of AI-powered tools for mathematical operations and holds potential applications in education.

In the context of a rapidly growing global population and significant climatic and environmental change, there is an urgent need to produce nutritious food in a sustainable manner. Some crops are underutilised in Europe, despite their suitability to local environments, viability for sustainable production and potential to improve diets. Rye ( Secale cereale ) has a long history of cultivation in Europe, yet is underutilised owing to complex historical, socio‐cultural, socio‐political, socio‐economic and agronomic factors. This paper explores an innovative, cross‐sectoral approach that harmonises existing datasets from archaeology, plant science, nutrition and policy, and establishes an interdisciplinary dialogue to tackle this challenge.

Background Poorly regulated and insufficiently maintained medical devices (MDs) carry high risk on safety and performance parameters impacting the clinical effectiveness and efficiency of patient diagnosis and treatment. As infant incubators are used as a form of fundamental healthcare support for the most sensitive population, prematurely born infants, special care mus be taken to ensure their proper functioning. This is done through a standardized process of post-market surveillance. Objective To address the issue of faulty infant incubators being undetected and used between yearly post-market surveillance, an automated system based on machine learning was developed for prediction of infant incubator performance status. Methods In total, 1997 samples were collected during the inspection process of infant incubator inspections performed by an ISO 17020 accredited laboratory at various healthcare institutions in Bosnia and Herzegovina. Various machine learning algorithms were considered, including Decision Tree (DT), Random Forest (RF), Naïve Bayes (NB) and Logistic Regression (LR) for the development of the automated system. Results The aforementioned algorithms were selected because of their ability to handle large datasets and their potential for achieving high prediction accuracy. The 0.93 AUC of Naïve Bayes indicates that it is overall stronger in predictive capabilities than decision tree and random forest which displayed superior accuracy in comparison to Naïve Bayes. Conclusion The results of this study demonstrate that machine learning algorithms can be effectively used to predict infant incubator performance status on the basis of measurements taken during post-market surveillance. Adoption of these automated systems based on artificial intelligence will help in overcoming challenges of ensuring quality of infant incubators that are already being used in healthcare institutions.

Three-dimensional (3D) voxel models have wide applications in fields such as robotics, medical imaging, autonomous navigation, and augmented reality. To address challenges of spatial sparsity and computational efficiency, this research proposes a voxel-based 3D convolutional neural network (3D CNN) integrated with a Transformer encoder for object classification on the ModelNet10 dataset. The 3D models are non-uniformly compressed into a 16×16×16 voxel grid, with compression factors incorporated as inputs to reduce sparsity. An in-house augmentation tool further enhances the dataset size and diversity by performing real-time voxel editing and labeling. The proposed method achieves an accuracy of 92.15% while utilizing only 184k trainable parameters, demonstrating an efficient and lightweight approach to object classification.

This paper presents a deep learning model that enables fast and accurate diagnosis of tuberculosis based on chest X-rays. The developed model uses convolutional neural network that enable the automatic classification of chest x-rays into one of two classes: Normal or Tuberculosis with a high degree of accuracy. The model achieved an accuracy of 97.55% on the test data set, indicating its potential to open new perspectives for medical professionals in establishing a tuberculosis diagnosis. This model can significantly speed up the diagnostic process, reducing the workload of medical workers and increasing their productivity in the fight against tuberculosis, one of the most common lung diseases.

This research presents a comparative analysis of modern voice cloning systems, focusing on their ability to generate high-quality speech from limited training data. The paper aims to demonstrate that carefully curated smaller datasets can produce superior results to larger, less structured datasets. The investigation of multiple state-of-the-art models, including Realtime Voice Cloning (RVC), Tortoise AI, Bark, and Coqui AI, establishes optimal data preparation protocols and identifies critical factors in training data quality, with particular emphasis on applications for the Montenegrin language and cultural preservation.

Artificial Intelligence is rapidly advancing the medical field by providing innovative disease diagnosis, treatment, and research approaches. This study explores the application of artificial intelligence in breast cancer diagnostics, focusing on using convolutional neural networks and deep learning to analyze mammographic images. ResNet152 and DenseNet121 models were used to classify malignant changes, achieving AUC scores exceeding 0.9, demonstrating their clinical utility. The research emphasizes how artificial intelligence can enhance screening efficiency, expedite diagnostic processes, and facilitate personalized treatment approaches. Ethical considerations, including patient safety and the transparency of artificial intelligence systems, were also analyzed. The findings underscore the potential of artificial intelligence to transform diagnostic procedures for breast cancer and highlight the importance of further research to integrate these technologies into clinical practice.

Viticulture in Montenegro faces significant challenges due to fragmented data management, limited access to high-resolution climate predictions, and the lack of systematic integration between stakeholders. This study addresses these issues by proposing a knowledge-driven system architecture that consolidates climate and phenology data, facilitates multi-level data sharing, and supports informed decision-making for sustainable vineyard management. Using Montenegro as a case study, the proposed decision support platform integrates data from Internet of Things-enabled climate pilots, existing databases, and predictive modeling tools to address limitations in existing datasets, such as low resolution and inaccurate downscaling methods, and to tackle the broader challenges posed by climate change, including shifting weather patterns and phenological cycles. The system architecture provides a framework for stakeholders, including researchers, winegrowers, and policymakers, to collaborate effectively, bridging the gap between localized data collection and high-level decision-making. The paper outlines the current state of viticulture in Montenegro and the EU, highlights the need for a systematic approach to data management, and details the benefits of such a system at various levels. The proposed platform architecture and implementation steps outlined in this study serve as a robust framework, offering valuable guidance for other countries seeking to establish similar systems to enhance the efficiency, sustainability, and resilience of their viticulture sectors. This research contributes to the broader understanding of knowledge-driven systems in precision agriculture and provides a scalable model for regions facing similar challenges.

Ready-to-eat (RTE) foods are the most common sources of Listeria monocytogenes transmission. Whole-genome sequencing (WGS) enhances the investigation of foodborne outbreaks by enabling the tracking of pathogen sources and the prediction of genetic traits related to virulence, stress, and antimicrobial resistance, which benefit food safety management. The aim of this study was to evaluate the efficacy of WGS in the typing of 16 L. monocytogenes strains isolated from refrigerated foods in Chile, highlighting its advantages in pathogen identification and the improvement of epidemiological surveillance and food safety. Using cgMLST, a cluster was identified comprising 2 strains with zero allele differences among the 16 strains evaluated. Ninety-four percent of the isolates (15/16) were serotype 1/2b, and 88% of them (14/16) were ST5. All strains shared identical virulence genes related to adhesion (ami, iap, lapB), stress resistance (clpCEP), invasion (aut, iapcwhA, inlAB, lpeA), toxin production (hly), and intracellular regulation (prfA), with only 13 strains exhibiting the bcrBC and qacJ gene, which confer resistance to quaternary ammonium. The pCFSAN010068_01 plasmids were prevalent, and insertion sequences (ISLs) and composite transposons (cns) were detected in 87.5% of the strains. The presence of various antibiotic resistance genes, along with resistance to thermal shocks and disinfectants, may provide L. monocytogenes ST5 strains with enhanced environmental resistance to the hygiene treatments used in the studied food production plant.

When we engage with fictions, we are, in effect, pretending to deal with reports of actual events. After all, numerous fictional works are explicitly designed to facilitate this kind of pretense. This was the prevailing understanding of fiction in both analytic philosophy and classical narratology for decades. However, there is a significant problem with this view: many fictional narratives routinely portray scenarios that could not possibly be the subject of anyone’s reporting. Currie’s 'mindless fictions' are one such example. This issue can be referred to as the problem of unreportability. The standard solution to this challenge has been to argue that fictions are not pretend reports, but rather direct authorial stipulations to imagine specific scenarios. This paper contends, however, that such an explanation fails to provide a satisfactory account of fiction. By drawing on Walton’s notion of ‘silly questions’, it instead argues for a revised version of the report model-one that doesn’t necessarily depend on ubiquitous fictional reporters.

Natural deep eutectic solvents (NaDES) were employed for the extraction of bilberry and green tea leaves. This study explored the incorporation of these NaDES extracts into various carrier systems: hydrogels, emulsions, and emulgels stabilized with hydroxyethyl cellulose or xanthan gum. The results demonstrated that, when combined with synthetic UV filters, the NaDES extracts significantly enhanced the SPF and improved the antioxidant properties of the formulation. Although NaDES extracts cannot fully replace synthetic UV filters (homosalate, ethylhexyl methoxycinnamate, and benzophenone-4), they can serve as effective UV boosters, significantly enhancing the SPFs of formulations containing UV filters. Hence, the SPF of the formulation could be improved without increasing the concentrations of synthetic filters. Moreover, NaDES extracts, unlike UV filters, significantly increased the antioxidant potential of the formulations. Among the carriers, hydrogels with xanthan gum and emulgels with hydroxyethyl cellulose achieved the highest SPFs when containing both NaDES extracts and synthetic filters. A texture analysis further revealed that the NaDES extracts positively impacted the mechanical properties of the formulations by increasing their cohesiveness, thus enhancing their physical stability under mechanical pressure. These findings pave the way for further research into NaDES-based formulations, including in vivo testing, to optimize and confirm their efficacy on human skin and validate NaDES extracts as eco-friendly ingredients in cosmetics, with antioxidant and UV boosting potential.

Introduction Enterococcus faecium is a widespread acid-lactic bacterium found in the environment, humans, and animal microbiota, and it also plays a role in the production of traditional food. However, the worldwide emergence of multidrug-resistant E. faecium strains represents a major public health threat and is the primary reason that the genus Enterococcus is not recommended for the Qualified Presumption of Safety (QPS) list of the European Food Safety Authority (EFSA), raising concerns about its presence in food products. Methods In this study, 39 E. faecium and 5 E. lactis isolates were obtained from artisanal brine cheeses and dry sausages, sourced from 21 different Montenegrin producers. The isolates were collected following the ISO 15214:1998 international method and processed for whole-genome sequencing (WGS). Results Genome analysis based on core genome multilocus sequence type (cgMLST) revealed a high diversity among isolates. Furthermore, the isolates carried antimicrobial resistance genes; the virulence genes acm, sgrA, and ecbA; the bacteriocin genes Enterolysin A, Enterocin A, Enterocin P, Duracin Q, Enterocin B, Bacteriocin 31, Enterocin EJ97, Sactipeptides, and Enterocin SEK4; the secondary metabolite genes T3PKS, cyclic lactone autoinducer, RiPP-like, and NRPS and a maximum of eight plasmids. Conclusion This study highlights the need for careful monitoring of E. faecium and E. lactis strains in food to ensure they do not pose any potential risks to consumer safety.

Background With the advancement of Artificial Intelligence (AI), clinical engineering has witnessed transformative opportunities, enabling predictive maintenance of medical devices, optimization of healthcare workflows, and personalized patient care. Respiratory equipment plays a vital role in modern healthcare, supporting patients with compromised or impaired respiratory capacities. However, ensuring the reliability and safety of these devices is crucial to prevent adverse events and ensure patient well-being. Objective This study aims to explore machine learning techniques to enhance predictive maintenance for mechanical ventilators. Method The dataset used for this study contains information about 1350 entries of mechanical ventilators, made by 15 different manufacturers and available in 30 distinct models. Different machine learning algorithms, including Logistic Regression, Decision Trees, Random Forest, K-nearest Neighbors, Support Vector Machines, Naive Bayes, and XG Boost are developed and tested in terms of their performance in predicting mechanical ventilator failures. Results The ensemble methods, particularly Random Forest and XGBoost, have proven to be more adept at handling the complexities of the dataset. The Decision Tree and Random Forest models both showed remarkable accuracies of approximately 0.993, while K-Nearest Neighbors (KNN) performed exceptionally with near perfect accuracy. Conclusion Adoption of automated systems based on artificial intelligence will help in overcoming challenges of ensuring quality of MDs that are already being used in healthcare institutions. Implementing machine learning-based predictive maintenance can significantly enhance the reliability of mechanical ventilators in healthcare settings.

The research presented in this paper explores the possibilities of optimizing solar energy usage and managing solar panels in hybrid energy systems that use both solar energy and the electrical grid. A model has been created to track time series data and solar radiation in order to predict the expected solar energy yield. The goal of this research is to enable users to plan their energy consumption and maximize savings through optimal use of solar panels. This process involves collecting weather data, optimizing energy usage, and proposing a system implementation that can be used by both companies and end users.