Search Results (169)

No systematic review has previously examined the application of large language models (LLMs) for generating impressions from radiology report findings. This study systematically reviews the performance of LLMs on this task and their associated evaluation methodologies. A search of seven electronic databases on 7 August 2025 identified 15 eligible papers (average quality score: 71.4%). These articles evaluated 35 LLMs, including 21 base models. The reported performance ranges were as follows: Recall-Oriented Understudy for Gisting Evaluation (ROUGE)-1, 35.9% (Generative Pre-Trained Transformer (GPT)-4) to 69.7% (Baichuan2-13B); ROUGE-2, 13.4% (Large Language Model Meta AI (Llama)) to 52.4% (Baichuan2-13B); and ROUGE-L, 16.5% (Chat General Language Model–Medical (ChatGLM-Med)) to 63.8% (finetuned Text-to-Text Transfer Transformer (T5)). The finetuned T5 consistently demonstrated high performance, based on Bidirectional Encoder Representations from Transformers Score (BERTScore): 89.2%; BiLingual Evaluation Understudy (BLEU)-1: 65.2%; BLEU-2: 57.9%; BLEU-3: 52.5%; BLEU-4: 48.3%; Metric for Evaluation of Translation with Explicit ORdering (METEOR): 38.1%; ROUGE-1: 59.9%; ROUGE-2: 50.9%; ROUGE-L: 63.8%; and subjective metrics (clinical usability: 4.5/5.0; completeness: 4.3/5.0; conciseness: 4.3/5.0; fluency: 4.4/5.0). These results, based on 132,043 computed tomography, echocardiography, magnetic resonance imaging, and X-ray reports, indicate its strong clinical potential for assisting radiologists in impression generation through supervised finetuning rather than prompting techniques used in closed-source LLMs. Full article

Generative models play a pivotal role in the field of medical imaging. This paper provides an extensive and scholarly review of the application of generative models in medical image creation and translation. In the creation aspect, the goal is to generate new images based on potential conditional variables, while in translation, the aim is to map images from one or more modalities to another, preserving semantic and informational content. The review begins with a thorough exploration of a diverse spectrum of generative models, including Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), Diffusion Models (DMs), and their respective variants. The paper then delves into an insightful analysis of the merits and demerits inherent to each model type. Subsequently, a comprehensive examination of tasks related to medical image creation and translation is undertaken. For the creation aspect, papers are classified based on downstream tasks such as image classification, segmentation, and others. In the translation facet, papers are classified according to the target modality. A chord diagram depicting medical image translation across modalities, including Magnetic Resonance Imaging (MRI), Computed Tomography (CT), Cone Beam CT (CBCT), X-ray radiography, Positron Emission Tomography (PET), and ultrasound imaging, is presented to illustrate the direction and relative quantity of previous studies. Additionally, the chord diagram of MRI image translation across contrast mechanisms is also provided. The final section offers a forward-looking perspective, outlining prospective avenues and implementation guidelines for future research endeavors. Full article

Deep learning models based on supervised learning rely heavily on large annotated datasets and particularly in the context of medical image segmentation, the requirement for pixel-level annotations makes the labeling process labor-intensive, time-consuming and expensive. To overcome these limitations, self-supervised learning (SSL) has emerged as a promising alternative that learns generalizable representations from unlabeled data; however, existing SSL frameworks often employ highly parameterized encoders that are computationally expensive and may lack robustness in label-scarce settings. In this work, we propose a scattering-based SSL framework that integrates Wavelet Scattering Networks (WSNs) and Parametric Scattering Networks (PSNs) into a Bootstrap Your Own Latent (BYOL) pretraining pipeline. By replacing the initial stages of the BYOL encoder with fixed or learnable scattering-based front-ends, the proposed method reduces the number of learnable parameters while embedding translation-invariant and small deformation-stable representations into the SSL pipeline. The pretrained encoders are transferred to a U-Net and fine-tuned for cardiac image segmentation on two datasets with different imaging modalities, namely, cardiac cine MRI (ACDC) and cardiac CT (CHD), under varying amounts of labeled data. Experimental results show that scattering-based SSL pretraining consistently improves segmentation performance over random initialization and ImageNet pretraining in low-label regimes, with particularly pronounced gains when only a few labeled patients are available. Notably, the PSN variant achieves improvements of 4.66% and 2.11% in average Dice score over standard BYOL with only 5 and 10 labeled patients, respectively, on the ACDC dataset. These results demonstrate that integrating mathematically grounded scattering representations into SSL pipelines provides a robust and data-efficient initialization strategy for cardiac image segmentation, particularly under limited annotation and domain shift. Full article

Background and Objectives: Academic point-of-care (POC) manufacturing enables the in-hospital design and production of patient-specific medical devices within certified environments, integrating clinical practice, engineering, and translational research. This model represents a new academic ecosystem that accelerates innovation while maintaining compliance with medical device regulations. Gregorio Marañón University Hospital has established one of the first ISO 13485-certified academic manufacturing facilities in Spain, providing on-site production of anatomical models, surgical guides, and custom implants for oral and maxillofacial surgery. This study presents a retrospective review of all devices produced between April 2017 and September 2025, analyzing their typology, materials, production parameters, and clinical applications. Materials and Methods: A descriptive, retrospective study was conducted on 442 3D-printed medical devices fabricated for oral and maxillofacial surgical cases. Recorded variables included device classification, indication, printing technology, material type, sterilization method, working and printing times, and clinical utility. Image segmentation and design were performed using 3D Slicer and Meshmixer. Manufacturing used fused deposition modeling (FDM) and stereolithography (SLA) technologies with PLA and biocompatible resin (Biomed Clear V1). Data were analyzed descriptively. Results: During the eight-year period, 442 devices were manufactured. Biomodels constituted the majority (approximately 68%), followed by surgical guides (20%) and patient-specific implants (7%). Trauma and oncology were the leading clinical indications, representing 45% and 33% of all devices, respectively. The orbital region was the most frequent anatomical site. FDM accounted for 63% of the printing technologies used, and PLA was the predominant material. The mean working time per device was 3.4 h and mean printing time 12.6 h. Most devices were applied to preoperative planning (59%) or intraoperative use (35%). Conclusions: Academic POC manufacturing offers a sustainable, clinically integrated model for translating digital workflows and additive manufacturing into daily surgical practice. The eight-year experience of Gregorio Marañón University Hospital demonstrates how academic production units can enhance surgical precision, accelerate innovation, and ensure regulatory compliance while promoting education and translational research in healthcare. Full article

Microwave sensing technology is rapidly advancing and increasingly finding its way into biomedical applications, promising significant improvements for medical care. Concurrently, the rise of artificial intelligence (AI) is enabling significant enhancements in the biomedical domain. Close scrutiny of the recent literature reveals intense activity in both fields, with particularly impactful outcomes deriving from the combined use of advanced microwave techniques and AI for biomedical monitoring. In this review, an up-to-date compilation, from the perspective of the authors, of the most significant works published on these topics in recent years is given, focusing on their integration and current challenges. With the objective of analyzing the current landscape, we survey and compare state-of-the-art biosensors and imaging systems at all healthcare levels, from outpatient contexts to specialized medical equipment and laboratory analysis tools. We also delve into the relevant applications of AI in medicine for processing microwave-derived data. As our core focus, we analyze the synergistic integration of AI in the design of microwave devices and the processing of the acquired data, which have shown notable performances, opening new avenues for compact, affordable, and multi-functional medical devices. We conclude by synthesizing the prevailing technical, algorithmic, and translational challenges that must be addressed to realize this potential. Full article

Background and Objectives: Dermatology relies on a complex terminology encompassing lesion types, distribution patterns, colors, and specialized sites such as hair and nails, while dermoscopy adds an additional descriptive framework, making interpretation subjective and challenging. Our study aims to evaluate the ability of a chatbot (Gemini 2) to generate dermatology descriptions across multiple languages and image types, and to assess the influence of prompt language on readability, completeness, and terminology consistency. Our research is based on the concept that non-English prompts are not mere translations of the English prompts but are independently generated texts that reflect medical and dermatological knowledge learned from non-English material used in the chatbot’s training. Materials and Methods: Five macroscopic and five dermoscopic images of common skin lesions were used. Images were uploaded to Gemini 2 with language-specific prompts requesting short paragraphs describing visible features and possible diagnoses. A total of 2400 outputs were analyzed for readability using LIX score and CLEAR (comprehensiveness, accuracy, evidence-based content, appropriateness, and relevance) assessment, while terminology consistency was evaluated via SNOMED CT mapping across English, French, German, and Greek outputs. Results: English and French descriptions were found to be harder to read and more sophisticated, while SNOMED CT mapping revealed the largest terminology mismatch in German and the smallest in French. English texts and macroscopic images achieved the highest accuracy, completeness, and readability based on CLEAR assessment, whereas dermoscopic images and non-English texts presented greater challenges. Conclusions: Overall, partial terminology inconsistencies and cross-lingual variations highlighted that the language of the prompt plays a critical role in shaping AI-generated dermatology descriptions. Full article

Ionic liquids (ILs) have emerged as multifunctional compounds with low volatility, high thermal stability, and tunable solvation capabilities, making them highly promising for biomedical applications. First explored in the late 1990s and early 2000s for enhancing the thermal stability of enzymes, antimicrobial agents, and controlled release systems, ILs have since gained significant attention in drug delivery, antimicrobial treatments, medical imaging, and biosensing. This review examines the diverse functions of ILs in contemporary therapeutics and diagnostics, highlighting their transformative capabilities in improving drug solubility, bioavailability, transdermal permeability, and pathogen inactivation. In drug delivery, ILs improve solubility of bioactive compounds, with several IL formulations achieving substantial solubility enhancements for poorly soluble drugs. Bio-ILs, in particular, show promise in enhancing drug delivery systems, such as improving transdermal permeability. ILs also exhibit significant antimicrobial and antiviral activity, offering new avenues for combating resistant pathogens. Despite their broad potential, challenges such as cytotoxicity, long-term metabolic effects, and the stability of ILs in physiological conditions persist. While much research has focused on their physicochemical properties, biological activity and in vivo studies are still underexplored. The future directions for ILs in biomedical applications include the development of bioengineered ILs and hybrid ILs, combining functional components like nanoparticles and polymers to create multifunctional materials. These ILs, derived from renewable resources, show great promise in personalized medicine and clinical applications. Further research is necessary to evaluate their pharmacokinetics, biodistribution, and long-term safety to fully realize their biomedical potential. This study emphasizes the potential of ILs to transform therapeutic and diagnostic technologies by highlighting present shortcomings and offering pathways for clinical translation, while also debating the need for continuous research to fully utilize their biomedical capabilities. Full article

Background and Objectives: Diffusion models, as a recent advancement in generative modeling, have become central to high-resolution image synthesis and reconstruction. Their rapid progress has notably shaped computer vision and health informatics, particularly by enhancing medical imaging and diagnostic workflows. However, despite these developments, researchers continue to face challenges due to the absence of a structured and comprehensive discussion on the use of diffusion models within clinical imaging. Methods: This systematic review investigates the application of diffusion models in medical imaging for diagnostic purposes. It provides an integrated overview of their underlying principles, major application areas, and existing research limitations. The review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines and included peer-reviewed studies published between 2013 and 2024. Studies were eligible if they employed diffusion models for diagnostic tasks in medical imaging; non-medical studies and those not involving diffusion-based methods were excluded. Searches were conducted across major scientific databases prior to the review. Risk of bias was assessed based on methodological rigor and reporting quality. Given the heterogeneity of study designs, a narrative synthesis approach was used. Results: A total of 68 studies met the inclusion criteria, spanning multiple imaging modalities and falling into eight major application categories: anomaly detection, classification, denoising, generation, reconstruction, segmentation, super-resolution, and image-to-image translation. Explainable AI components were present in 22.06% of the studies, clinician engagement in 57.35%, and real-time implementation in 10.30%. Overall, the findings highlight the strong diagnostic potential of diffusion models but also emphasize the variability in reporting standards, methodological inconsistencies, and the limited validation in real-world clinical settings. Conclusions: Diffusion models offer significant promise for diagnostic imaging, yet their reliable clinical deployment requires advances in explainability, clinician integration, and real-time performance. This review identifies twelve key research directions that can guide future developments and support the translation of diffusion-based approaches into routine medical practice. Full article

Endometrial hyperplasia (EH) comprises a spectrum of abnormal proliferative changes in the endometrium, ranging from benign glandular overgrowth to lesions with substantial malignant potential. The importance of risk stratification and early identification is highlighted by the growing recognition of EH as a precursor to endometrial cancer. The main causes of EH, according to epidemiological research, include obesity, polycystic ovarian syndrome (PCOS), metabolic dysfunction, and extended exposure to unopposed estrogen. Emerging molecular markers, histological analysis, and imaging are all necessary for a proper diagnosis of EH because it might appear with vague clinical symptoms such as irregular uterine bleeding. Surgical intervention or progestin therapy are two possible management techniques for EH, depending on the lesion’s intricacy and the patient’s medical history, including fertility issues. Personalized therapy techniques and recent developments in molecular profiling have the potential to enhance patient outcomes by matching treatment to tumor biology and individual risk profiles. This review highlights the translational potential of molecular insights while synthesizing the most recent data on the epidemiology, risk factors, diagnostic techniques, and therapy of EH. A deeper comprehension of these elements is necessary to maximize treatment results and stop the development of endometrial cancer. Full article

Online adaptive radiotherapy (ART) is rapidly transforming clinical radiation oncology by enabling adaptation of treatment plans based on patient-specific anatomical and biological changes. However, most medical physics training programs lack structured education in ART. To address this critical gap, the Medical Physics Adaptive Radiotherapy (MPART) Fellowship was established at our center to train post-residency or practicing physicists in advanced adaptive technologies and workflows. The MPART Fellowship is a two-year program that provides immersive, platform-specific training in CBCT-guided (Varian Ethos), MR-guided (Elekta Unity), and PET-guided (RefleXion X1) radiotherapy. Fellows undergo modular clinical rotations, hands-on training, and dedicated research projects. The curriculum incorporates competencies in imaging, contouring, online planning, quality assurance, and team-based decision-making. Evaluation is based on the Accreditation Council for Graduate Medical Education competency domains and includes milestone tracking, mentor reviews, and structured presentations. The fellowship attracted applicants from both domestic and international institutions, reflecting strong demand for formal ART training. Out of 22 applications, two fellows have been successfully recruited into the program since 2024. Fellows actively participate in all phases of adaptive workflows and are expected to function at near-attending levels by the second year of their training. Each fellow also leads at least one translational or operational research project aimed at improving ART delivery. Fellows contribute to clinical coverage and lead developmental projects, resulting in presentations and publications at the national and international levels. The MPART Fellowship addresses a vital educational need by equipping medical physicists with the advanced competencies necessary for implementing and leading ART. This program offers a replicable framework for other institutions seeking to advance precision radiation therapy through structured post-residency training in adaptive radiotherapy. As this fellowship program is still in its early phase of establishment, the primary goal of this paper is to introduce the structure, framework, and implementation model of the program. Comprehensive outcome analyses—such as quantitative assessments, fellow feedback, and longitudinal competency evaluations—will be incorporated in future work as additional cohorts complete training. Full article

Background/Objectives: Since the invention of MRI, analytical methods for MRI data have continuously evolved. In recent years, the rapid development of artificial intelligence has transformed MRI data analysis—from functional MRI (fMRI) techniques to deep learning-based image segmentation, and from traditional machine learning to radiomics for clinical applications. Methods: This review provides a succinct summary of recent progress in fMRI and structural MRI analysis. The discussed techniques include fMRI, quantitative MRI (qMRI) methods such as T1 and T2 relaxation time mapping, and proton density imaging. Approaches for diffusion, perfusion, and the Dixon method are also described. Furthermore, studies published between 2012 and 2025 on MRI radiomics were reviewed. Different neural network architectures related to radiomics-based segmentation are compared and discussed. Results: A major trend in both fMRI and MRI analysis is the increasing use of quantitative methods, which enable better cross-study comparison and reproducibility. Deep learning remains to progress rapidly in MRI research, particularly in segmentation tasks, with new loss functions and network architectures developed to improve performance. These methods are expected to undergo further optimization and find broader applications in clinical practice. Conclusions: Despite substantial progress, challenges remain in standardization, validation, and clinical translation. Continued efforts are necessary before these advanced analytical techniques can be fully integrated into routine medical practice. Full article

Background: Chronic Obstructive Pulmonary Disease (COPD) represents a significant global health burden, characterized by complex diagnostic and management challenges. Artificial Intelligence (AI) presents a powerful opportunity to enhance clinical decision-making and improve patient outcomes by leveraging complex health data. Objectives: This scoping review aims to systematically map the existing literature on AI applications in COPD. The primary objective is to identify, categorize, and summarize research into three key domains: (1) Diagnosis, (2) Clinical Symptoms, and (3) Clinical Outcomes. Methods: A scoping review was conducted following the Arksey and O’Malley framework. A comprehensive search of major scientific databases, including PubMed, Scopus, IEEE Xplore, and Google Scholar, was performed. The Population–Concept–Context (PCC) criteria included patients with COPD (Population), the use of AI (Concept), and applications in healthcare settings (Context). A global search strategy was employed with no geographic restrictions. Studies were included if they were original research articles published in English. The extracted data were charted and classified into the three predefined categories. Results: A total of 120 studies representing global distribution were included. Most datasets originated from Asia (predominantly China and India) and Europe (notably Spain and the UK), followed by North America (USA and Canada). There was a notable scarcity of data from South America and Africa. The findings indicate a strong trend towards the use of deep learning (DL), particularly Convolutional Neural Networks (CNNs) for medical imaging, and tree-based machine learning (ML) models like CatBoost for clinical data. The most common data types were electronic health records, chest CT scans, and audio recordings. While diagnostic applications are well-established and report high accuracy, research into symptom analysis and phenotype identification is an emerging area. Key gaps were identified in the lack of prospective validation and clinical implementation studies. Conclusions: Current evidence shows that AI offers promising applications for COPD diagnosis, outcome prediction, and symptom analysis, but most reported models remain at an early stage of maturity due to methodological limitations and limited external validation. Future research should prioritize rigorous clinical evaluation, the development of explainable and trustworthy AI systems, and the creation of standardized, multi-modal datasets to support reliable and safe translation of these technologies into routine practice. Full article

With the rapid development of medical imaging technology, the early diagnosis and treatment of heart disease have been significantly improved. Cardiac CT (Computed Tomography) and ultrasound images are often used in combination to provide more comprehensive information on cardiac structure and function due to their respective advantages and limitations. However, due to the significant differences in imaging principles, resolutions, and viewing angles between these two imaging modalities, how to effectively register cardiac CT and ultrasound images has become an important research topic in imaging and clinical applications. This article summarizes the research progress of cardiac CT and ultrasound image registration, and analyzes the existing registration methods and their advantages and disadvantages. Firstly, this article summarizes traditional registration methods based on image intensity, feature points, and regions, and explores the application of rigid and non-rigid registration algorithms. Secondly, in view of common challenges in cardiac CT and ultrasound image registration, such as image noise, deformation, and differences in imaging time, this article discusses the recent advances in multimodal registration technology in cardiac imaging and forecasts the potential of deep learning methods in registration. In addition, this article also evaluates the application effects and limitations of these methods in clinical practice, and finally looks forward to the future development direction of cardiac image registration technology, especially its potential applications in personalized medicine and real-time monitoring. Through a comprehensive review of the current research status of cardiac CT and ultrasound image registration, this article provides a systematic theoretical framework for researchers in related fields and provides a reference for future technological breakthroughs and clinical translation. Full article

Background/Objectives: The healthcare data sector is experiencing unprecedented growth, fueled by advances in genomics, medical imaging, and wearable devices. The convergence of universal data standards now provides the common ground needed to translate this data into medical advances. However, a significant implementation gap persists, preventing effective deployment in routine clinical practice, particularly in specialized areas like diabetic foot care. Methods: This paper examines the opportunities presented by modern data methodologies to bridge this gap, contextualized within diabetic foot care, where the paramount goals are patient well-being, tissue preservation, and amputation prevention. Results: The analysis indicates that the synergy of interoperable data and advanced management tools is poised to fundamentally transform healthcare delivery. Interdisciplinary collaboration is identified as the foundational element enabling the timely, coordinated, and evidence-based interventions necessary to achieve critical clinical objectives. Conclusions: The pivotal challenge has shifted from technological capability to effective implementation. Leveraging modern data methodologies is essential for translating potential into tangible improvements in diabetic foot outcomes. In this context, collaborative data management must be recognized as a critical treatment modality itself. Here, “data is tissue”; it must be managed with the same urgency and care to enable success. Full article

Background: The clinical imperative to reduce patient ionizing radiation exposure during diagnosis and treatment planning necessitates robust, high-fidelity synthetic imaging solutions. Current cross-modal synthesis techniques, primarily based on GANs and deterministic CNNs, exhibit instability and critical errors in modeling high-contrast tissues, thereby hindering their reliability for safety-critical applications such as radiotherapy. Objectives: Our primary objective was to develop a stable, high accuracy framework for 3D Magnetic Resonance Imaging (MRI) to Computed Tomography (CT) synthesis capable of generating clinically equivalent synthetic CTs (sCTs) across multiple anatomical sites. Methods: We introduce a novel 3D Latent Diffusion Model (3DLDM) that operates in a compressed latent space, mitigating the computational burden of 3D diffusion while leveraging the stability of the denoising objective. Results: Across the Head & Neck, Thorax, and Abdomen, the 3DLDM achieved a Mean Absolute Error (MAE) of 56.44 Hounsfield Units (HU). This result demonstrates a significant 3.63% reduction in overall error compared to the strongest adversarial baseline, CycleGAN (MAE = 60.07 HU, p < 0.05), a 10.76% reduction compared to NNUNet (MAE = 67.20 HU, p < 0.01), and a 20.79% reduction compared to the transformer-based SwinUNeTr (MAE = 77.23 HU, p < 0.0001). The model also achieved the highest structural similarity (SSIM = 0.885 ± 0.031), significantly exceeding SwinUNeTr (p < 0.0001), NNUNet (p < 0.01), and Pix2Pix (p < 0.0001). Likewise, the 3D-LDM achieved the highest peak signal-to-noise ratio (PSNR = 29.73 ± 1.60 dB), with statistically significant gains over CycleGAN (p < 0.01), NNUNet (p < 0.001), and SwinUNeTr (p < 0.0001). Conclusions: This work validates a scalable, accurate approach for volumetric synthesis, positioning the 3DLDM to enable MR-only radiotherapy planning and accelerate radiation-free multi-modal imaging in the clinic. Full article