Search Results (241)

Background/Objectives: Post-stroke disability is a prevalent complication in patients who have experienced a stroke. It is imperative that patients suffering from associated disability be hospitalised in rehabilitation wards, with a view to minimising their disability. The primary objective of the present study is to analyse the direct medical costs associated with the rehabilitation of patients with post-stroke disability who are admitted to a rehabilitation clinic for the first time. Methods: This retrospective study was conducted in a public hospital in Romania between January 2021 and December 2024. Patient information was retrieved from the hospital database and included the following: socio-demographic and clinical characteristics; disability score assessed using the modified Rankin scale (mRS); number of days of hospitalisation; and direct medical costs related to hospitalisation. Results: A total of 584 patients were included in this study. The average age was 68.04 years, 82% had suffered an ischaemic stroke, and 18% had suffered a haemorrhagic stroke. The mRS disability scores for ischaemic stroke were 2 (28.54%); 3 (24.79%); 4 (30.41%); and 5 (16.25%). The mRS scores for haemorrhagic stroke were 4 (33.65%); 5 (29.80%); 3 (20.19%); and 2 (16.34%). Hypertension was present in 80% of patients. The average length of hospital stay was 12.44 days. The total cost of hospitalisation per patient averaged RON 5295.33 thousand (approximately EUR 1031). Pearson’s correlation indicates a statistically significant positive association between higher mRS disability scores and higher hospitalisation costs (p < 0.001). Conclusions: The financial burden imposed on healthcare systems in Romania by medical expenses related to the rehabilitation of patients with post-stroke disability is significant. It is imperative to implement measures that will reduce the financial burden associated with hospitalising these patients and minimise the duration of their hospital stay. Full article

Background: During postoperative recovery, patients and their caregivers often lack crucial information, leading to numerous repetitive inquiries that burden healthcare providers. Traditional discharge materials, including paper handouts and patient portals, are often static, overwhelming, or underutilized, leading to patient overwhelm and contributing to unnecessary ER visits and overall healthcare overutilization. Conversational chatbots offer a solution, but Natural Language Processing (NLP) systems are often inflexible and limited in understanding, while powerful Large Language Models (LLMs) are prone to generating “hallucinations”. Objective: To combine the deterministic framework of traditional NLP with the probabilistic capabilities of LLMs, we developed the AI Virtual Assistant (AIVA) Platform. This system utilizes a retrieval-augmented generation (RAG) architecture, integrating Gemini 2.0 Flash with a medically verified knowledge base via Google Vertex AI, to safely deliver dynamic, patient-facing postoperative guidance grounded in validated clinical content. Methods: The AIVA Platform was evaluated through 750 simulated patient interactions derived from 250 unique postoperative queries across 20 high-frequency recovery domains. Three blinded physician reviewers assessed formal system performance, evaluating classification metrics (accuracy, precision, recall, F1-score), relevance (SSI Index), completeness, and consistency (5-point Likert scale). Safety guardrails were tested with 120 out-of-scope queries and 30 emergency escalation scenarios. Additionally, groundedness, fluency, and readability were assessed using automated LLM metrics. Results: The system achieved 98.4% classification accuracy (precision 1.0, recall 0.98, F1-score 0.9899). Physician reviews showed high completeness (4.83/5), consistency (4.49/5), and relevance (SSI Index 2.68/3). Safety guardrails successfully identified 100% of out-of-scope and escalation scenarios. Groundedness evaluations demonstrated strong context precision (0.951), recall (0.910), and faithfulness (0.956), with 95.6% verification agreement. While fluency and semantic alignment were high (BERTScore F1 0.9013, ROUGE-1 0.8377), readability was 11th-grade level (Flesch–Kincaid 46.34). Conclusion: The simulated testing demonstrated strong technical accuracy, safety, and clinical relevance in simulated postoperative care. Its architecture effectively balances flexibility and safety, addressing key limitations of standalone NLP and LLMs. While readability remains a challenge, these findings establish a solid foundation, demonstrating readiness for clinical trials and real-world testing within surgical care pathways. Full article

The accurate diagnosis of Hip Osteoarthritis (HOA) and the prediction of Total Hip Arthroplasty (THA) outcomes are crucial for reliable decision-making on treatment and rehabilitation strategies. Gait analysis (GA) is commonly employed for gait disorder examination in clinical settings, but it is still limited due to the massive data size and accuracy problems. A Machine Learning (ML) methodology has seen rapid growth in the past decade, but its development in the context of HOA and THA GA has not been previously examined. Thus, the novel contribution of this review is the evaluation of the current state of ML frameworks for the analysis of HOA and post-THA gaits. Five databases, namely PubMed, Embase, IEEE Xplore, ACM Digital Library, and Scopus, were searched in accordance with the PRISMA framework. Relevant publications published until May 2025 were retrieved, and information on reliability, applicability, and interpretability were extracted for quality assessment. Out of the 759 publications initially considered, 19 studies were selected, with 14 articles focused on classification and 5 articles on outcome prediction. Eight classification studies utilized kinematic features, while four outcome prediction articles utilized spatiotemporal parameters and mostly focused on post-THA gaits. The reported accuracy ranges between 70 and 100%, with the support vector machine (SVM) as the most frequently utilized ML algorithm. Scarce datasets, small sample sizes, and limited design description were the main hindrances revealed in our quality assessment. Nevertheless, this review demonstrated the recent developments in the utilization of ML techniques and evidently improved applicability through a consensus on the important gait features for HOA and post-THA gait analysis. Reliability and interpretability are still major concerns before ML models become widely accepted by medical practitioners. Future research should consider dataset quality, transparent validation protocol, model interpretability, and results’ explainability. Full article

Health misinformation across digital platforms has emerged as a critical fast-growing challenge to global public health, undermining trust in science and contributing to vaccine hesitancy, treatment refusal and heightened health risks. In response, this study introduces Impact, a novel simulation framework that integrates agent-based modeling (ABM) with large language model (LLM) integration and retrieval-augmented generation (RAG) to evaluate and optimize health communication strategies in complex online environments. By modeling virtual populations characterized by demographic, psychosocial, and emotional attributes, embedded within network structures that replicate the dynamics of digital platforms, the framework captures how individuals perceive, interpret and propagate both factual and misleading health messages. Messages are enriched with evidence from authoritative medical sources and iteratively refined through sentiment analysis and comparative testing, allowing the proactive pre-evaluation of diverse communication framings. Results demonstrate that misinformation spreads more rapidly than factual content, but that corrective strategies, particularly empathetic and context-sensitive messages delivered through trusted peers, can mitigate polarization, enhance institutional trust and sustain long-term acceptance of evidence-based information. These findings underscore the importance of adaptive, data-driven approaches to health communication and highlight the potential of simulation-based methods to inform scalable interventions capable of strengthening resilience against misinformation in digitally connected societies. Full article

The development of new technologies, improved by advances in artificial intelligence, has enabled the creation of a new generation of applications in different scenarios. In medical systems, adopting AI-driven solutions has brought new possibilities, but their effective impacts still need further investigation. In this context, a chatbot prototype trained with large language models (LLMs) was developed using data from the Santa Catarina Telemedicine and Telehealth System (STT) Dermatology module. The system adapts Llama 3 8B via supervised Fine-tuning with QLoRA on a proprietary, domain-specific dataset (33 input-output pairs). Although it achieved 100% Fluency and 89.74% Coherence, Factual Correctness remained low (43.59%), highlighting the limitations of training LLMs on small datasets. In addition to G-Eval metrics, we conducted expert human validation, encompassing both quantitative and qualitative aspects. This low factual score indicates that a retrieval-augmented generation (RAG) mechanism is essential for robust information retrieval, which we outline as a primary direction for future work. This approach enabled a more in-depth analysis of a real-world telemedicine environment, highlighting both the practical challenges and the benefits of implementing LLMs in complex systems, such as those used in telemedicine. Full article

Background: Influenza can cause severe complications, especially in patients with specific risk factors or comorbidities associated with poor outcomes. Some patients are at increased risk of a complicated disease course, including secondary infections, ICU admission, and the need for mechanical ventilation. The first post–COVID-19 seasonal influenza season placed a substantial burden on Dutch ICUs. This study investigates the disease course and outcomes of ICU patients with influenza. Methods: A retrospective influenza registry study was conducted across 34 Dutch ICUs, including patients aged 18 and older admitted to the ICU with a positive influenza RT-PCR test, between 1 November 2023 and 17 March 2024. Data on demographic information, medical history, clinical symptoms, laboratory and imaging results, parameters of mechanical ventilation, additional treatments, length of hospital stay, and mortality was retrieved from the electronic patient record. Results: A total of 498 patients were included in the study. The median age was 64 (IQR: 55–72) years and 58.8% of the patients were men. The most common comorbidities were cardiovascular disease (34.1%), chronic obstructive pulmonary disease (COPD) (31.5%), and diabetes (22.3%). Bacterial co-infections were present in 37.6% of the patients. Invasive mechanical ventilation (IMV) was necessary in 46.0% of patients, 38.0% of those requiring IMV were treated in prone position. A substantial mortality rate was observed, with an ICU mortality rate of 21.9% and an additional hospital mortality rate of 5.2%. Conclusion: This study described the characteristics and course of disease of all patients with laboratory-confirmed influenza infection admitted to one of the 34 participating Dutch ICUs between November 2023 and March 2024. The major findings of this study are the substantial mortality rate, a high proportion of patients with bacterial co-infections, and a significant percentage of patients requiring IMV and prone position ventilation. Finally, patients without comorbidities that were admitted to the ICU with an influenza virus infection showed severe disease parameters but had a lower mortality than patients with comorbidities. Full article

Accessing reliable medical information is a major challenge for healthcare professionals due to limited accessibility to real-time medical data sources. The study’s objectives are maximization of response accuracy with minimal latency and enhancement of the model’s interpretability. An explainable bilingual medical-question-answering system (MQAS) is introduced to improve accessibility and trust in healthcare information retrieval. Using knowledge-aware networks (KANs), retrieval augmented generation (RAG), and linked open data (LOD), a synthetic bilingual dataset is generated. Through the application of a synthetic dataset and Bayesian optimization HyperBand (BOHB)-based hyperparameter optimization, the performance of GPT-Neo and RoBERTa models is fine-tuned. The outcomes of GPT-Neo and RoBERTa are ensembled using the weighted majority voting approach, while Shapley Additive ExPlanation (SHAP) value provides interpretability and transparency. The proposed model is trained and evaluated using diverse medical-question-answering datasets, demonstrating superior performance over baseline models. It achieves a generalization accuracy of 90.58%, an F1-score of 89.62%, and a BLEU score of 0.80 with a low inference time of 3.4 s per query. The findings underscore the model’s potential in delivering accurate, bilingual, and explainable medical responses. This study establishes a foundation for building multilingual healthcare information systems, promoting inclusive and equitable access to medical information. Full article

Background: Biopsy remains the gold standard for characterizing breast cancer, but it is invasive, costly, and may not fully capture tumor heterogeneity. Advances in artificial intelligence (AI) now allow for the extraction of biological and clinical information from medical images, raising the possibility of using imaging as a non-invasive alternative. Methods: A semi-systematic review was conducted to identify AI-based approaches applied to mammography (MM) and breast tomosynthesis (BT) for tumor subtyping, staging, and prognosis. A PubMed search retrieved 1091 articles, of which 81 studies met inclusion criteria (63 MM, 18 BT). Studies were analyzed by clinical target, modality, AI pipeline, number of cases, dataset type, and performance metrics (AUC, accuracy, or C-index). Results: Most studies focused on tumor subtyping, particularly receptor status and molecular classification. Contrast-enhanced spectral mammography (CESM) was frequently used in radiomics pipelines, while end-to-end deep learning (DL) approaches were increasingly applied to MM. Deep models achieved strong performance for ER/PR and HER2 status prediction, especially in large datasets. Fewer studies addressed staging or prognosis, but promising results were obtained for axillary lymph node (ALN) metastasis and pathological complete response (pCR). Multimodal and longitudinal approaches—especially those combining MM or BT with MRI or ultrasound—show improved accuracy but remain rare. Public datasets were used in only a minority of studies, limiting reproducibility. Conclusions: AI models can predict key tumor characteristics directly from MM and BT, showing promise as non-invasive tools to complement or even replace biopsy. However, challenges remain in terms of generalizability, external validation, and clinical integration. Future work should prioritize standardized annotations, larger multicentric datasets, and integration of histological or transcriptomic validation to ensure robustness and real-world applicability. Full article

The development of clinical decision support systems for treatment planning and monitoring faces significant challenges, such as high labor intensity, integration complexities, lack of universality, and insufficient consideration of individual patient characteristics. This paper presents an innovative approach to overcoming these limitations, based on the creation of a specialized software toolkit. The key feature of the proposed approach is the use of a hybrid decision-making mechanism that integrates knowledge-based reasoning and case-based reasoning. For knowledge representation, a universal generalized ontology was developed, capable of modeling information about different treatment modalities (pharmacological, rehabilitative, surgical) while remaining independent of any specific medical specialty. This enabled the creation of a unified decision-making algorithm. For case retrieval, a combined method was proposed. The toolkit is being actively used on the IACPaaS platform to develop treatment planning systems across various medical domains, demonstrating its practical applicability and effectiveness. Full article

Background/Objectives: Non-specific low back pain (LBP), a highly prevalent musculoskeletal condition, may be associated with overuse of radiological imaging, despite clinical guidelines restricting its use to cases with suspected serious pathology. This study investigated demographic, clinical, and physiotherapy-related factors influencing radiological imaging use in patients with non-specific LBP. Methods: A retrospective cross-sectional study included 179 non-specific LBP patients from an outpatient physiotherapy clinic in Saudi Arabia. Patient data were anonymized and retrieved from electronic health records, including demographic, clinical, physiotherapy and imaging information. Independent variables included patient demographics, non-specific LBP characteristics, physiotherapy engagement, and pain-related outcomes. Descriptive, inferential, and multiple linear regression analyses were conducted to identify predictors of radiological imaging. Results: Among the total study sample (n = 179), 159 (88.8%) patients underwent radiological imaging, primarily X-ray (32.4%) and Magnetic Resonance Imaging (8.4%); 48.0% received multiple imaging modalities. Significant predictors of imaging use included gender (p < 0.001), higher body mass index (BMI) (p = 0.012), greater physiotherapist experience (p = 0.019), and presence of comorbidities (p = 0.023). Non-specific LBP medication use was negatively associated with imaging (p = 0.032). Physiotherapy engagement and pain-related outcomes showed no significant impact on imaging use. Conclusions: Gender, BMI, physiotherapist experience, and comorbidities could influence radiological imaging use in non-specific LBP patients. These findings highlight potential biases in imaging referral patterns and reinforce the need for adherence to evidence-based guidelines to prevent unnecessary imaging, reduce healthcare costs, and enhance patient care. Full article

Large-scale medical visual question answering (MedVQA) datasets are critical for training and deploying vision–language models (VLMs) in radiology. Ideally, such datasets should be automatically constructed from routine radiology reports and their corresponding images. However, no existing method directly links free-text findings to the most relevant 2D slices in volumetric computed tomography (CT) scans. To address this gap, a contrastive language–image pre-training (CLIP)-based key slice selection framework is proposed, which matches each sentence to its most informative CT slice via text–image similarity. This experiment demonstrates that models pre-trained in the medical domain already achieve competitive slice retrieval accuracy and that fine-tuning them on a small dual-supervised dataset that imparts both lesion- and organ-level awareness yields further gains. In particular, the best-performing model (fine-tuned BiomedCLIP) achieved a Top-1 accuracy of 51.7% for lesion-aware slice retrieval, representing a 20-point improvement over baseline CLIP, and was accepted by radiologists in 56.3% of cases. By automating the report-to-slice alignment, the proposed method facilitates scalable, clinically realistic construction of MedVQA resources. Full article

Practicing medical research based on the best evidence is gaining increased value and popularity among most medical societies in the current era. Good clinical practice (GCP) is internationally recognized as the scientific and ethical standard for the design, conduct, performance, auditing, recording, analysis, and reporting of clinical trials involving human subjects. GCP ensures the accuracy and credibility of trial while safeguarding the rights, integrity, and confidentiality of participants. Adherence to GCP facilitates the generation of high-quality studies that can be incorporated in Evidence-Based Medicine (EBM). The clinical practice of EBM seeks to integrate robust medical literature into daily medical practice. This process involves systematically searching for high-quality evidence, critically appraising the retrieved literature, applying sound clinical principles and finally evaluating the efficacy of the chosen approach. Although EBM has been evaluated in many resource settings, it has not been addressed sufficiently in the field of Pediatrics and more specifically in indigenous populations. In this review, we briefly explain the EBM approach and its applications in Pediatrics, in order to help physicians care for young subjects more efficiently by integrating the best available information into their routine clinical practice. Also, the basic good practice principles for conducting clinical trials in children and adolescents are highlighted, emphasizing the importance of applying high ethical principles in this vulnerable population. Full article

Each year an estimated 440,000 medical errors occur in the U.S., of which 38% are a direct result of patient matching errors. As patients seek care in medical facilities, their records are often dispersed. Health Information Exchanges (HIEs) strive to retrieve and consolidate these records and as such, accurate matching of patient data becomes a critical prerequisite. Artificial intelligence (AI) is increasingly being seen as a potential solution to this vexing challenge. We present findings from an exploratory field study involving interviews with 27 HIE executives across the U.S. on tensions they are sensing and balancing in incorporating AI in patient matching processes. Our analysis of data from the interviews reveals, on the one hand, significant optimism regarding AI’s capacity to improve matching processes, and on the other, concerns due to the risks associated with algorithmic biases, uncertainties regarding AI-based decision-making, and implementation hurdles such as costs, the need for specialized talent, and insufficient datasets for training AI models. We conceptualize this dialectical tension in the form of a grounded theory framework on Cautious AI Optimism. Full article

Electronic Health Records (EHRs) have enhanced access to medical information but have also introduced challenges for healthcare providers, such as increased documentation workload and reduced face-to-face interaction with patients. To mitigate these issues, we propose RAGMed, a Retrieval-Augmented Generation (RAG)-based AI assistant designed to deliver automated and clinically grounded responses to frequently asked patient questions. This system combines a vector database for semantic retrieval with the generative capabilities of a large language model to provide accurate, reliable answers without requiring direct physician involvement. In addition to patient-facing support, the assistant facilitates appointment scheduling and assists clinicians by summarizing clinical notes, thereby streamlining healthcare workflows. Additionally, to evaluate the influence of retrieval quality on overall system performance, we compare two embedding models, gte-large and all-MiniLM-L6-v2, using real-world medical queries. The models are assessed within the RAG-Triad Framework, focusing on context relevance, answer relevance, and factual groundedness. The results indicate that gte-large, owing to its higher-dimensional embeddings, retrieves more informative context, resulting in more accurate and trustworthy responses. These findings underscore the importance of not only the potential of incorporating RAG-based systems to alleviate physician workload and enhance the efficiency and accessibility of healthcare delivery but also the dimensionality of models used to generate embeddings, as this directly influences the relevance, accuracy, and contextual understanding of retrieved information. This prototype is intended for the retrieval-augmented answering of medical FAQs and general informational queries, and is not designed for diagnostic use or treatment recommendations without professional validation. Full article

Background/Objectives: With the rapid development of smart healthcare, medical records have shifted from a disease-centered to a patient-centered approach. However, traditional formats, such as narratives and tables, often make it challenging for physicians to quickly grasp a patient’s condition within a limited timeframe, potentially leading to diagnostic errors and a decline in the quality of care. Recently, advances in information visualization and 3D technology have led many medical institutions to employ charts and graphs or use 3D simulations of organs to support clinical practice and education. However, few have integrated 3D models into medical records for use during physician consultations. Methods: This study presents the development and evaluation of a novel web-based 3D EMR system that integrates real-time ICD-10 diagnostic code mapping with interactive 3D human body models, enabling physicians to visualize patient-specific anatomical and diagnostic information in a dynamic and context-aware manner. Results: We employed the System Usability Scale (SUS) to evaluate the system’s usability, conducting a satisfaction survey. Results from the survey indicate that participants rated the system highly in terms of ease of use, satisfaction, and efficiency, with an average SUS score of 70.42, reflecting usability between moderate and good. Comparative evaluations and future expansion plans are also discussed. Conclusions: These findings demonstrate that integrating a 3D human model into the medical record retrieval process significantly improves visualization and interactivity, meeting the needs of healthcare professionals and enhancing both their efficiency and patient satisfaction. Full article