The Visual-inertial navigation system (VINS) integrates visual and inertial sensors, providing advanced navigation capabilities. Recent improvements in these sensors have made VINS widely applicable in areas such as robotic navigation and autonomous driving, due to its complementary functionality and decreasing sensor costs. This paper reviews the latest developments in VINS, introducing visual simultaneous localization and mapping and its role in VINS. Key technologies, including direct and indirect methods for image processing in visual odometry and Inertial Measurement Unit preintegration for robot motion estimation, are discussed. Both filter-based and optimization-based state estimation methods in VINS are examined. Filter-based methods, like the Extended Kalman Filter, offer real-time state updates for attitude tracking and map construction, while optimization-based methods focus on minimizing reprojection or other error metrics to improve localization accuracy and robustness. The application of dynamic simultaneous localization and mapping, which addresses dynamic objects in complex environments, is also explored. This paper summarizes current research challenges and proposes future directions in the field of dynamic simultaneous localization and mapping.

Purpose: This study investigates the effects of varying inner diameter of a vascular clamp made from an Mg–Nd–Zn–Zr alloy on its functional performance. The primary objectives are to optimize the clamp’s structure, assess its performance across different inner diameters, and ultimately determine the optimal configuration. Methods: We developed a V-shaped vascular clamp equipped with a locking mechanism and transverse teeth. The study involved comparing vascular clamps with various inner diameters (0.35 mm, 0.4 mm, 0.5 mm and 0.6 mm, denoted as R0.35, R0.4, R0.5 and R0.6, respectively), achieved by modifying the clamp design. Finite element analysis simulated the closure process of these clamps, both with and without blood vessels, to analyze stress and strain distribution. Subsequently, we manufactured a clamp with the optimized design and conducted performance evaluations, including a closing strength test and an in vitro immersion test. Results: Among the tested vascular clamps, the R0.5 clamp demonstrated the lowest strain (0.50798) and minimal stress on blood vessels (0.7629 MPa). Notably, the R0.5 clamp remained intact during clamping fracture experiments and demonstrated a maximum closing force of 334.98±15.4 mmHg. Regarding corrosion resistance, the clamped position showed a higher corrosion rate (0.179±0.00551 mg.cm^-2.day^-1) compared to the open clamp (0.161±0.00306 mg.cm^-2.day^-1). Conclusion: The R0.5 clamp demonstrated superior performance in finite element analysis, showing effective vascular closure, strong clamping force, and uniform corrosion behavior. Overall, these results highlight its potential as an effective tool for vascular closure.

Photoacoustic imaging (PAI) technology, characterized by its high resolution, minimal biological impact, and high  sensitivity, has become a cornerstone in biomedical research. Its application spans various domains, showing  significant promise for disease diagnosis. Currently, the majority of PAI research is conducted using animal models, with human clinical applications still in early development. This paper reviews the fundamental principles of  PAI and explores its use in animal brain imaging studies. It addresses the current challenges and limitations of  the technology and evaluates the potential for extending these techniques to human cerebral imaging. PAI offers  substantial benefits for diagnosing neurological disorders, and its adaptation for human brain studies is crucial for advancing our understanding of neuropathogenesis, improving early disease detection, and monitoring treatment effectiveness. Continued advancements in PAI are expected to not only augment its role in neuroscience research but also establish it as a valuable tool in clinical diagnostics.

With the continuous advancement in medical device technology, minimally invasive surgery has become the cornerstone of modern surgical practices. At the forefront of this evolution is the fusion of medical endoscopes with high-frequency electrosurgical instruments, now a mainstream approach in minimally invasive surgeries, driving the development of innovative surgical procedures. This paper aims to provide an in-depth understanding of the principles of electrosurgical units, with a particular focus on standard procedures in gastrointestinal endoscopic electrosurgery. The goal of this review is to provide a more profound and comprehensive insight into endoscopic electrosurgery for medical practitioners and patients. Through the comprehensive study, it is anticipated to serve as a guide and reference for improving surgical outcomes, reducing patient discomfort, and simplifying the tasks of healthcare professionals.

Advanced endoscopic therapeutic techniques, such as endoscopic mucosal resection, endoscopic submucosal dissection, and third-space endoscopy, have transformed digestive endoscopy into a minimally invasive surgical platform. This transformation, nonetheless, poses a major challenge, namely the safe and efficient closure of the postoperative gastrointestinal (GI) defects that arise, as inadequately closed defects may lead to serious postoperative complications. This review provides a comprehensive overview of the historical background, the basic principles of closure, and the technical changes of the mainstream devices used for closure of GI defects, viewed mostly from the perspective of medical devices and engineering design. We have divided the most significant closure devices currently in use into endoscopic clipping devices and endoscopic suturing devices, and we have outlined their respective historical development and working principles. In addition, this review has identified and explained advanced clip-based closure methods that employ multiple through-the-scope clips (TTSCs) or are combined with auxiliary devices to overcome the inherent limitations of single-TTSC use. This review aims to provide endoscopists with a deeper understanding of the existing closure devices so as to  facilitate their optimal clinical use, and to offer insights that may help in the development of next-generation closure technologies.

Pulmonary function testing (PFT) is a critical tool for assessing respiratory health and managing diseases such as chronic obstructive pulmonary disease (COPD) and asthma. Leveraging its powerful natural language processing and big data analytics capabilities, Generative pre-trained transformer (GPT) can efficiently integrate multidimensional patient data to generate precise diagnostic reports and treatment recommendations, significantly enhancing interpretation efficiency and supporting clinical decision-making. Its applications encompass monitoring and early warning, comprehensive analysis, diagnostic assessment, and personalized treatment plan formulation. Particularly in managing chronic airway diseases like COPD and asthma, GPT provides individualized advice by tracking pulmonary function data, symptoms, and medication responses in real-time; it assists in dynamic assessment and prognosis prediction for interstitial lung diseases; and in the surgical domain, it supports the evaluation of operative tolerance and optimization of perioperative management. Furthermore, GPT can predict drug tolerance issues, offer suggestions for medication adjustments, and provide comprehensive health interventions incorporating lifestyle and environmental factors. This technology also extends remote medical services, mitigating disparities in healthcare resource distribution by enhancing service accessibility through data integration and conversion, intelligent analysis and reporting, and remote consultations. Implementation requires addressing challenges such as privacy protection, content accuracy, data preprocessing, user experience, and system stability. These can be addressed through measures like encryption technologies, constructing medical knowledge bases, developing data conversion modules, and optimizing interface design to ensure the secure and reliable application of GPT in pulmonary function testing.

Key Words: generative pre-trained transformer; artificial intelligence; chronic obstructive pulmonary diseases; asthma

In order to cope with the challenges and inherent limitations in the development of medical GPT technology, the author suggests a comprehensive and in-depth innovation strategy, which requires a refined reshaping of every link from data acquisition to system operation and maintenance. Data collection is no longer just a quantitative accumulation but a qualitative leap, which means carefully selecting from a vast amount of medical information to ensure that each piece of data is highly representative, accurate, and usable. This process requires not only the support of advanced technical means, but also the in-depth participation of medical experts to achieve accurate data screening and value mining. In the selection of the pedestal model, the traditional simple question and answer framework should be abandoned, and the possibility of building an expert digital human doppelganger should be explored. This transformation allows patients to receive more personalized and professional medical consultation services as if they were directly facing experienced doctors, which greatly improves the interactive experience and trust. At the same time, in order to ensure the security and accuracy of medical information, it is recommended to apply an AI-based intelligent quality control mechanism to replace the blind reliance in the past, and strictly control the quality through a combination of automatic review by algorithm and manual review. In addition, the training, evaluation and optimization of models should also pay more attention to the integration of practical experience. On the basis of evidence-based medicine, it advocates the integration of the clinical wisdom and experience of big doctors into the model, so that medical GPT technology can not only provide patients with more accurate and individualized diagnosis and treatment suggestions based on the latest scientific research results, but also combine with clinical practice. In short, it is necessary to realize the four major transformations from data cleaning to selection, from simple consultation to expert clone, from blind trust to quality control, and from simple evidence-based to combined with the experience of doctors.

Key Words: artificial intelligence; generative pretrained transformer; medical generative pretrained transformer; natural language processing; open evidence

This study presents an intelligent system for patient education based on large language model (LLMs) and multi agent collaboration, achieving end-to-end process management through five specialized agents (Ask Agent, Assessment Agen, Advice Agent, Arrangement Agent, and Assistance Agent). The system not only improves treatment efficiency, but also provides more precise personalized treatment plans for complex cases, demonstrating significant value in enhancing patient education quality.

Key Words:  large language model; multi-agent system; artificial intelligence; patient education

Obesity/overweight has become a complex metabolic disease involving multiple systems, and the traditional model of relying on outpatient follow-up and short-term weight loss is difficult to solve the problems of insufficient prediction, extensive intervention, poor adherence and unequal resources. In recent years, artificial intelligence (AI/ML), large language models (LLM/GPT), Internet of Things (IoT), metaverse and digital humans have been introduced into the whole chain of weight management: in the prevention stage, multimodal risk models and wearable devices achieve early identification and behavior warning of high-risk groups, and metaverse scenarios improve health education and participation; In the diagnostic stage, AI supports obesity phenotype reconstruction, automatic body composition analysis, and complication “red light signal” recognition, promoting weight control from “BMI-centric” to “phenotype and complication-centric”; In the treatment and rehabilitation stage, AI-assisted accurate benefit prediction of weight loss drugs such as GLP-1, personalized push of digital therapy and behavioral coaching, full-process risk assessment of metabolic surgery, and promotion of functional recovery and anti-obesity through VR/GDTx and digital twins. At the management level, a comprehensive platform based on cloud-edge-end architecture and embedded with 5P medical concepts has been formed, and medical GPT/BAIMGPT connects doctors, patients and managers as a “weight loss digital human expert”. The main challenges include data quality and generalization, fairness and privacy protection, clinical process and payment model adaptation, and GPT security regulation. Looking forward to the future, the deep integration of AI, digital twins, and the metaverse is expected to achieve systematic improvement from simple “weight loss” to “metabolic health and self-management ability”.

Key Words: weight control; AI; Internet of Things; medical GPT; metaverse

Cardiopulmonary resuscitation (CPR) is a critical life-saving skill, and its widespread adoption is closely associated with the survival rate of out-of-hospital cardiac arrest. However, traditional CPR training remains several limitations. With the rapid development of artificial intelligence, virtual reality (VR), augmented reality, and metaverse-related technologies, immersive training has emerged as a promising innovation in medical education, particularly in CPR instruction. This article reviews the research progress and challenges of VR-based CPR system, and provides ideas for future technology development.

Keywords: cardiopulmonary resuscitation; virtual reality; metaverse; medical education; immersive training; artificial intelligence; skill transfer