Frontiers in medical technology最新文献

Background: Magnetic Resonance Imaging (MRI) and ultrasound are central to tumour diagnosis and treatment planning. Although Deep learning (DL) models achieve strong prediction performance, high computational demand and limited explainability can hinder clinical adoption. Common post hoc Explainable Artificial Intelligence (XAI) methods namely Grad-CAM, LIME, and SHAP often yield fragmented or anatomically misaligned saliency maps.

Methods: We propose SpikeNet, a hybrid framework that combines Convolutional Neural Networks (CNNs) for spatial feature encoding with Spiking Neural Networks (SNNs)for efficient, event driven processing. SpikeNet includes a native saliency module that produces explanations during inference. We also introduce XAlign, a metric that quantifies alignment between explanations and expert tumour annotations by integrating regional concentration, boundary adherence, and dispersion penalties. Evaluation follows patient level cross validation on TCGA-LGG (MRI, 22 folds) and BUSI (ultrasound, 5 folds), with slice level predictions aggregated to patient level decisions and BUSI treated as a three class task. We report per image latency and throughput alongside accuracy, precision, recall, F1, AUROC, and AUPRC.

Results: SpikeNet achieved high prediction performance with tight variability across folds. On TCGA-LGG it reached $97.12\pm 0.63\%$ accuracy and $97.43\pm 0.60\%$ F1; on BUSI it reached $98.23\pm 0.58\%$ accuracy and $98.32\pm 0.50\%$ F1. Patient level AUROC and AUPRC with 95% confidence intervals further support these findings. On a single NVIDIA RTX 3090 with batch size 16 and FP32 precision, per image latency was about 31 ms and throughput about 32 images per second, with the same settings applied to all baselines. Using XAlign, SpikeNet produced explanations with higher alignment than Grad-CAM, LIME, and SHAP on both datasets. Dataset level statistics, paired tests, and sensitivity analyses over XAlign weights and explanation parameters confirmed robustness.

Conclusion: SpikeNet delivers accurate, low latency, and explainable analysis for MRI and ultrasound by unifying CNN based spatial encoding, sparse spiking computation, and native explanations. The XAlign metric provides a clinically oriented assessment of explanation fidelity and supports consistent comparison across methods. These results indicate the potential of SpikeNet and XAlign for trustworthy and efficient clinical decision support.

Purpose: Magnetic resonance imaging (MRI) during cardiopulmonary bypass is hindered by the incompatibility of conventional heart-lung machines, which contain metallic components that interfere with the MRI environment. This study evaluates the hemolytic performance of three MR-conditional blood pump prototypes-roller, non-occlusive roller, and centrifugal-designed for use during neonatal surgery.

Materials and methods: Each pump was tested using acid-citrate dextrose-stabilized bovine blood at a neonatal-relevant flow rate of 1 L/min. Due to limitations of the setup, a low pressure head of 10 mmHg was applied uniformly across experiments. Hemolysis was assessed using normalized index of hemolysis, and a linear mixed-effects model was applied to account for experimental variability.

Results: The roller pump showed the lowest hemolysis (1.84 ± 1.90 mg/100 L). The centrifugal pump showed the highest (8.43 ± 1.63 mg/100 L), alongside mechanical leakage. Random effects (SD = 2.07) indicated moderate inter-experimental variability.

Conclusion: While all prototypes performed comparably to standard references under controlled conditions, further testing at physiological pressure levels and stricter adherence to ASTM F1841 is necessary for broader validation.

Background: Artificial intelligence (AI)-based models can augment clinical decision-making, including prediction, diagnosis, and treatment, in all aspects of medicine.

Research questions: The current systematic review aims to provide a summary of existing data about the role of machine learning (ML) techniques in predicting in-hospital cardiac arrest, life-threatening ventricular arrhythmias, and respiratory arrest.

Methods: The study was conducted in compliance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) framework. PubMed, Embase, and Web of Science without any restriction were searched to extract relevant manuscripts until October 20, 2023. Additionally, the reference list of all potential studies was searched to identify further relevant articles. Original publications were regarded as eligible if they only recruited adult patients (≥18 years of age), employed AI/ML algorithms for predicting cardiac arrest, life-threatening ventricular arrhythmias, and respiratory arrest in the setting of critical care, used data gathered from wards with critically ill patients (ICUs, cardiac ICUs, and emergency departments), and were published in English. The following information was extracted: first author, journal, ward, sample size, performance and features of ML and conventional models, and outcomes.

Results: ML algorithms have been used for cardiac arrest prediction using easily obtained variables as inputs. ML algorithms showed promising results (AUC 0.73-0.96) in predicting cardiac arrest in different settings, including critically ill ICU patients, patients in the emergency department and patients with sepsis, they demonstrated variable performance (AUC 0.54-0.94) in predicting respiratory arrest in COVID-19 patients, as well as other clinical settings.

Conclusion: ML algorithms have shown promising results in predicting in-hospital cardiac and respiratory arrest using readily available clinical data. These algorithms may enhance early identification of high risk patients and support timely interventions, thereby reducing mortality and morbidity rates. However, the prospective validation of these algorithms and their integration into clinical workflows need further exploration.

Introduction: KineFeet, a depth camera-based web tool for analyzing functional foot kinematics, was developed and tested in this study. The program was optimized for usability, affordability, and clinical relevance through an iterative design and development process.

Methods: The Azure Kinect DK camera records and analyzes sagittal and frontal plane foot movements in real time. A usability-focused study was created. Five physiatrists tested the KineFeet prototype for its ability to assess foot kinematics. Performance was measured by task completion success, error rate, and time. The System Usability Scale (SUS) measured user satisfaction. Quality assessments were also obtained through semi-structured interviews.

Results: Participants achieved an average success rate of 96.29%, with an error rate of 0.074% and an average completion time of 10 min 11 s. Time-Based Efficiency (TBE) showed that user performance (0.0442 tasks/s) was 1.21 times slower than expert user performance (0.05348 tasks/s). SUS yielded an average score of 66.5, indicating a good level of satisfaction and user acceptance.

Conclusion: KineFeet represents a promising innovation in assessing functional foot kinematics. The system demonstrated strong usability in preliminary testing and holds potential for broader clinical adoption following further development.

Central nervous system (CNS) infections caused by pathogens such as HIV, Herpes simplex virus, Cryptococcus neoformans, and Toxoplasma gondii remain among the most difficult to treat due to the physiological barrier posed by the blood-brain barrier (BBB), pathogen latency, and systemic toxicity associated with conventional therapies. Exosome-based delivery systems are becoming a game-changing platform that can solve these therapeutic problems using their natural biocompatibility, minimal immunogenicity, and capacity to cross the BBB. This review current developments in exosome engineering that aim to make brain-targeted therapy for neuroinfectious illnesses more selective and effective. Much focus is on new molecular methods like pathogen-specific ligand display, aptamer conjugation, lipid modification, and click-chemistry-based surface functionalisation. These methods make it possible to target diseased areas of the brain precisely. Exosomes can also carry therapeutic payloads, such as anti-viral and antifungal drugs, gene editing tools like CRISPR/Cas9 and siRNA, and more. This makes them helpful in changing pathogens' persistence and the host's immunological responses. The paper tackle problems with translation, such as biodistribution, immunogenicity, GMP production, and regulatory issues. Future possibilities like synthetic exosomes, combinatory medicines, and delivery design that uses AI. The combination of nanotechnology, molecular biology, and infectious disease therapies shows that exosome engineering offers a new way to meet the clinical needs that are not satisfied in treating CNS infections.

Introduction: Early and accurate detection of diabetes-related foot ulcers (DFU) that may become chronic is essential to prevent long-term disability, amputation, and mortality. Various non-invasive imaging techniques have been developed to detect and monitor DFU progression, but none have yet been widely adopted in clinical practice. This review summarizes current advancements in non-invasive image techniques for DFU wound healing prediction and identifies research directions to support clinical translation.

Methods: A systematic, multi-disciplinary review was conducted focusing on three imaging methods: photographic, hyperspectral, and thermal imaging. Articles published between July 2014 and July 2024 were searched across five databases: PubMed, Scopus, CINAHL, Embase, and Web of Science. The search was limited to English-language, peer-reviewed journal articles. The review followed PRISMA guidelines and applied the CASP quality appraisal tool.

Results: The initial search identified 2,937 articles, of which 22 studies met the inclusion criteria, including 17 original studies (9 medical and 8 engineering) on DFU healing prediction using imaging techniques and 5 relevant review articles.

Discussion: Each imaging method offers specific benefits and faces unique limitations: photographic imaging is user-friendly but lighting-sensitive; thermal imaging reflects inflammation but requires multimodal integration; hyperspectral imaging provides biochemical insight but is costly and less portable. Visual and thermal imaging, in particular, demonstrate strong potential for early and real-time prediction when combined with machine learning/deep learning. These methods offer portability, ease of use, and potential for automated analysis on a single device, making them suitable for clinical and community settings. However, challenges such as standardization and integration complexity remain. Continued research with larger datasets and improved validation is needed to enhance clinical readiness.

In biopharmaceutical production involving cells, cell-derived products, or tissues, the cleaning of surfaces that come into direct or indirect contact with the product is currently performed mostly by hand as the initial step in decontamination. This manual approach leads to production inefficiencies, reduced reproducibility of decontamination processes, and product losses. In food production, automated processes are preferred for the decontamination of interior product contact surfaces. This article studies the feasibility of adapting the automated cleaning-in-place concepts used in the food industry to biopharmaceutical production. The focus is on spray cleaning processes and validation by cleaning simulation. An existing automated cell production platform is used as a case study for validation. The results indicate that modifying an existing platform to support cleaning-in-place presents significant challenges. However, the article outlines general design guidelines for developing new biopharmaceutical production platforms that can accommodate automated cleaning.

Background: This study aims to develop a machine learning model to predict the 30-day mortality risk of hospitalized COVID-19 patients while leveraging federated learning to enhance data privacy and expand the model's applicability. Additionally, SHapley Additive exPlanations (SHAP) values were utilized to assess the impact of comorbidities on mortality.

Methods: A retrospective analysis was conducted on 6,321 clinical records of hospitalized COVID-19 patients between January 2021 and October 2022. After excluding cases involving patients under 18 years of age and non-Omicron infections, a total of 4,081 records were analyzed. Key features included three demographic data, six vital signs at admission, and 79 underlying comorbidities. Four machine learning models were compared, including Lasso, Random Forest, XGBoost, and TabNet, with XGBoost demonstrating superior performance. Federated learning was implemented to enable collaborative model training across multiple medical institutions while maintaining data security. SHAP values were applied to interpret the contribution of each comorbidity to the model's predictions.

Results: A subset of 2,156 records from the Taipei branch was used to evaluate model performance. XGBoost achieved the highest AUC of 0.96 and a sensitivity of 0.94. Two versions of the XGBoost model were trained: one incorporating vital signs, suitable for emergency room applications where patients come in with unstable vital signs, and another excluding vital signs, optimized for outpatient settings where we encounter patients with multiple comorbidities. After implementing federated learning, the AUC of the Taipei cohort decreased to 0.90, while the performance of other cohorts improved to meet the required standards. SHAP analysis identified comorbidities including diabetes mellitus, cerebrovascular disease, and chronic lung disease to have a neutral or even protective association with 30-day mortality.

Conclusion: XGBoost outperformed other models making it a viable tool for both emergency and outpatient settings. The study underscores the importance of chronic disease assessment in predicting COVID-19 mortality, revealing some comorbidities such as diabetes mellitus, cerebrovascular disease and chronic lung disease to have protective association with 30-day mortality. These findings suggest potential refinements in current treatment guidelines, particularly concerning high-risk conditions. The integration of federated learning further enhances the model's clinical applicability while preserving patient privacy.

Objective: To explore the feasibility of enhancing the accuracy of percutaneous lung puncture through matching the respiratory activity of the thorax and performing puncture under the guidance of a laser angle guider.

Methods: A retrospective case-control study was adopted. Collected data of patients with pulmonary nodules undergoing puncture. They were categorized into the conventional puncture group (Con group), the laser guidance group (Laser group), and the thoracic respiratory activity matching group (Ram group) based on whether the puncture was guided by the laser angle guider and the application of the thorax respiratory activity matching technique.

Results: 277 patients were included: 96 in the Con group, 93 in the Laser group, and 88 in the Ram group. There were no statistically significant differences in the puncture purpose, age, gender, BMI, maximum diameter of pulmonary nodules, location of pulmonary nodules, and distance from the skin at the puncture point among the three groups (P > 0.05). Nevertheless, key outcomes showed the Ram group had better results than the Laser group, which were better than the Con group: the rate of reaching the predetermined position on the first puncture (67.05% vs. 37.63% vs. 23.96%), number of CT scans (3.66 ± 1.06 vs. 4.09 ± 1.05 vs. 4.50 ± 1.08 times), and procedure time (23.05 ± 13.89 vs. 28.83 ± 13.78 vs. 35.14 ± 14.20 min) (all P < 0.05). Complication rates were sequentially lower (7.95% vs. 16.13% vs. 26.04%; P = 0.146).

Conclusion: Puncture by matching the thoracic respiratory activity and under the guidance of a laser angle guider can effectively improve the accuracy of percutaneous lung puncture, reduce complications. Furthermore, the procedure is straightforward, warranting further evaluation in larger, prospective studies.

Clinical trial registration: Chinese Clinical Trial Registry (ChiCTR), identifier (ChiCTR2300069384).