This empirical research study mainly evaluates the effectiveness of three frequency-based incremental type of feature extraction and machine learning classification and prediction approaches, i.e. LSTM, RNN, and AdaBoostM2 in detecting driver’s fatigue while monitoring driver’s vigilance using EEG data from the SEED-VIG dataset. This dataset comprises EEG signaling data samples which are categorized into three states of driver’s alertness, i.e. drowsiness, tiredness and awake state. The study aims to determine which model can most accurately detect driver fatigue, a critical factor in preventing road accidents. By starting from the lowest and second lowest frequency bands in EEG data for feature selection, and through incremental feature extraction and deep learning approaches in detecting the EEG data, AdaBoostM2 outperformed the other two prediction models. It achieved the highest prediction accuracy of 97.03%, compared with 95.85% and 94.28% for LSTM and RNN respectively. These results showed that AdaBoostM2 is a superior tool for real-time driver vigilance monitoring.
Transcranial direct current stimulation (tDCS) is a non-invasive brain stimulation technique that has shown potential in enhancing motor function through modulating brain activity. In this study, we developed a real-time closed-loop tDCS-EEG system aimed at supporting motor rehabilitation by adapting tDCS current amplitude based on ongoing EEG signals. Our system can adjust tDCS parameters within seconds and analyze EEG data in 40ms time windows. Implemented within a motor imagery (MI) paradigm, the system was tested with 8 healthy subjects, achieving a response time of 200ms for closed-loop regulation and a modulation period of 10 seconds. Results demonstrated significant effects on motor-related EEG features, suggesting the system’s feasibility in modulating motor-related brain activity for practical applications in the rehabilitation field.
This research aims to comparatively investigate, evaluate and develop various prediction models by using Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest, Adaptively generated Association Ruled Pre-pruned or AgARPp AdaBoostM2, AgARPp Gradient Boosting, and AgARPp XGBoost 1 to predict and classify the three different outcomes of diabetes, i.e. diabetes, pre-diabetes and non-diabetes. This is based on various factors such as demographic, clinical, lifestyle factors and attributes of individuals. The primary objective is to identify the most accurate prediction algorithm that can be used in the healthcare industry to assist healthcare professionals in the early and more accurate diagnosis of various stages of diabetes. The dataset used is the online diabetes dataset. Besides ANN, SVM and Random Forest, the rest of AdaBoostM2, Gradient Boosting and XGBoost are preprocessed using Adaptively generated Association Ruled Pre-pruning techniques. Results after applying the six prediction models were comprehensively analyzed, compared and studied. These prediction models were evaluated based on their respective evaluation metrics, mainly accuracy, precision, recall and F1-score. Among the results of all the six evaluated prediction models, AgARPp AdaBoostM2 showed the best prediction accuracy result of 98.06% when this model was being implemented.
Hyperspectral images contain spatial and spectral information that can be used for pathological condition identification, disease detection, or resection margin assessment during surgery. Developing hyperspectral detection systems that are convenient for clinical examination and surgical site use can help improve medical care quality. Collaborative robots (Cobot) with force sensors are highly safe and can keep humans and robotic arms safe when operating near humans. Nowadays, Cobot is used increasingly in clinical practice fields. In this study, a push-broom hyperspectral image acquisition system, driven by Cobot, targeted for medical applications in clinical practice fields was developed. A push-broom hyperspectral imager driven by a Cobot with multiple degrees of freedom can have high flexibility in the scan direction, position, and mode. The hyperspectral imager, used as the end effector of Cobot, can be easily switched to scan the patient's body parts under different body postures (e.g., standing, sitting, laying) with varying modes of scan (linear, swing, and orbiting). Using linear mode, this Cobot driving hyperspectral imager can obtain hyperspectral image data similar to those used in the industrial sorting line. The swing mode can let the push-broom scan have minimum imager/robotic arm motion when scanning. This feature allows a high spectral resolution push-broom imager to scan body parts in confined space. The orbiting mode can reduce the edge effects when scanning body parts with significant curvature. For patients with mobility impairment, hemiplegia, or under surgical operation, the scanning system can accommodate the patient’s postures to scan the patient's body parts. The system developed by this project will enable hyperspectral technology to be more widely used in clinical medical fields. This system can be used to assist doctors in performing visual assessments of surface symptoms and enable hyperspectral technology to be more commonly used in medical institutions.
Ultrasound shear wave elastography imaging (USWE) is a new technology that quantifies the mechanical properties of tissue, such as elasticity and viscosity. It has been used to detect tumours in many organs such as the liver, breast, prostate, etc. The simulation of ultrasound elastography is critical for future advancements in the field, particularly, computer-aided diagnostics using elastography methods. Most USWE simulation studies in the literature assume that the tissue is an elastic medium, therefore neglecting the attenuation behaviour of tissues. In this work, an accurate simulation of shear wave propagation in a viscoelastic tissue medium has been done using the k-Wave toolbox in MATLAB. Subsequently, the shear wave speed has been calculated using the time to peak (TTP) method to analyze the propagation of the generated shear waves. The result shows the calculated shear wave speed of the simulated wave has less than 2% error. In addition, the decrease in amplitude of the shear waves has also been observed with the propagation of shear waves, which shows the shear wave attenuation due to the viscoelasticity of the material. The wave dispersion analysis on the generated shear wave from 200 Hz to 2000 Hz shows an increase in the phase velocities with the frequency.
We propose a multi-stage deep learning approach for heartbeat-wise detection of atrial fibrillation (AF) from electrocardiogram (ECG) signals. Our method combines sample-level classification with beat-level aggregation, culminating in a QRS complex-focused analysis that aligns closely with clinical ECG interpretation practices. The model utilizes a deep neural network architecture featuring 1-D convolutional layers for feature extraction and bidirectional long short-term memory (BiLSTM) layers for capturing temporal dependencies. We trained and validated the model using carefully selected records from the MIT-BIH Arrhythmia Database. Our approach demonstrates exceptional performance in beat-wise AF detection, particularly in the final QRS-focused stage. On the independent test set, we achieved an overall accuracy of 99.80% for AF detection. The model showed near-perfect precision, sensitivity, and specificity for both AF and normal beat classification, with AF detection showing 1.0 precision, 0.9981 sensitivity, and 1.0 specificity. The key innovation of our work lies in the progressive refinement of the classification process, from sample (time-stamp) level to beat-wise, and ultimately focusing on the QRS complex region. This stepwise approach significantly enhances detection accuracy. By concentrating on the QRS complex, our method captures the most relevant information for distinguishing between normal and AF beats, improving both accuracy and physiological relevance of the classification process. This work has significant implications for improving the accuracy of AF diagnosis in clinical settings, potentially leading to more timely interventions and improved patient outcomes.
Many current dynamic hand orthoses use single degree of freedom joints, such as hinge joints. Consequently, these orthoses can only partially replicate the complex, multi-axis range of motion of the hand. To overcome this limitation, one approach is to use prestressed compliant structures as the basis for orthoses. In order to be able to map the dynamic influences on the orthosis as well as to simulate everyday use, this article provides an overview of this orthosis concept and the most important aspects in the development of such an orthosis including the results of the modal analysis. Based on these theoretical investigations, the presented methodological approach can be used to develop initial prototypes of tensegrity-based hand orthoses.
In modern society, the prevalence of sedentary lifestyles, driven by office work and the widespread use of technology such as smartphones, televisions, and computers, has led to insufficient physical activity and associated health issues like obesity. According to the World Health Organization (WHO), 31% of adults and 80% of adolescents fail to meet the recommended levels of physical activity. Many individuals also lack awareness of their daily exercise levels.To address this problem, we propose a transformer-based model that leverages wearable sensor data to monitor and classify whole-body movements, helping users determine whether they are physically inactive. The model was evaluated on four benchmark datasets: MobiAct, UniMiB SHAR, USC-HAD, and UCI HAR, achieving F1 Scores of 92.01%, 93.40%, 89.00%, and 87.89%, respectively, outperforming baseline models. These results demonstrate the model’s effectiveness in promoting physical activity awareness and addressing sedentary behaviors.
This study aimed to investigate the movement of upper anterior teeth during clear aligner treatment using the finite element method, comparing two approaches: the interference fit method and the initial stress method. The study found that the results of the vector plot displacement in both the crown and root regions showed that the two methods caused the anchorage teeth to move in different directions, but the movement in the activated teeth was consistent. and when comparing the displacement values, it was found that smaller teeth experienced greater movement than larger teeth. As for the activated tooth, the tooth movement across the methods was consistent, with the displacement values showing little difference. The study also indicated that the tooth movement observed during clear aligner treatment aligns with clinical issues, particularly anchorage loss, which is commonly encountered during orthodontic treatment.
Alzheimer's disease (AD) is a neurodegenerative disease that is mainly characterised by an insidious onset and subtle clinical symptoms, making it difficult to diagnose by conventional means. Electroencephalography (EEG) has been widely used to detect Alzheimer's disease as an important tool for assessing and aiding the diagnosis of brain disorders. Accurate diagnosis is essential to prevent progression from early cognitive impairment to AD and to provide early treatment for AD patients. Most classification methods focus only on the time or frequency domain, but ignore the relationship between the two as well as the spatial information of the channels. To address this problem, this paper proposes a feature fusion two-branch parallel network (FFDBPNet) for AD diagnosis, which consists of a convolutional neural network (CNN) and a branch of a long short-term memory (LSTM) recurrent neural network. First, the EEG data are pre-processed and transformed into time-frequency spectrograms, which are dimensionally reduced and fed into the FFDBPNet network. The CNN extracts the time-frequency spectral spatial features of multi-band EEG signals, and the LSTM extracts the time-series features of EEG signals. In addition, a flattening layer is added after the pooling layer to further extract intermediate features to improve the generalisation ability of the model. Finally, feature fusion is performed to obtain the classification results. The classification accuracy of the proposed model for AD EEG signals reaches 96.09%, the sensitivity is 98.04%, and the specificity is 94.14%, which exceeds the traditional classification methods. The experimental results show that the feature fusion model proposed in this paper can extract effective time-frequency features and spatial features from EEG for classification.
This study investigates the mechanical behavior of glass fiber composite components bonded with an adhesive layer, offering insights for the design of biomedical implants and other engineering applications. Two surface geometries were analyzed: a smooth surface and a grooved surface, to evaluate the effect of geometric design on stress distribution and overall joint strength. Numerical simulations were performed using the ABAQUS software, employing the porcelain layer approach to simulate a 18-layer composite structure accurately. The adhesive was modeled with both elastic and plastic properties, and failure analysis was conducted using the maximum stress criterion to identify and eliminate overstressed elements. In addition to the numerical analysis, recommendations for experimental validation are provided to enhance the reliability of the findings. Furthermore, the study discusses the influence of material diversity, environmental conditions (such as humidity and temperature), and complex surface geometries on joint performance. The results demonstrate that grooved surfaces promote more uniform stress distribution and improve joint strength, which is crucial for enhancing the longevity and stability of adhesively bonded components in biomedical and industrial applications.
In this paper we propose a new robust water-marking scheme for medical images stored in a distributed database with a hyperbolic structure. Our approach is to first generate a watermark from the patient information and the statistical parameters of the image. The watermark is then inserted into the image and fragmented before being distributed to different nodes. The repartition is carried out using the hyper-catadioptric projection model in order to guarantee its robustness. The receiver extracts the watermark from the image and look for watermark fragments stored in the nodes before proceeding with a comparison. Simulation results clearly show that any attempt to modify the image or the watermark will be detected. The asynchronous stream cryptography technique has been proposed to restore data in the event of an attack. The implementation of the hyper-catadioptric model enabled us to adapt our watermarking scheme to the structure of the distributed database. Finally, formal analysis and simulations allow us to state that our system is chaotic and therefore robust.
With the widespread adoption of RF energy in electrosurgery, electrosurgery has become an essential element in most surgical procedures. In particular, monopolar electrodes, which are indispensable in most open surgeries, have been the subject of extensive studies focusing on electrode shape, material, coating, as well as RF generator control and temperature and/or impedance sensing. In electrosurgery, which delivers thermal energy to tissue for cutting and coagulation, heat represents both the most critical benefit and the greatest risk. The high operating temperatures of electrodes (exceeding 250 degrees C), designed for efficiency as surgical tools, are known to cause excessive thermal damage to tissues and implanted devices. This paper proposes a novel blade-type monopolar electrode technology that employs highly focused dielectric heating as its energy transfer mechanism, in contrast to conventional monopolar electrode devices, which rely on ohmic loss and heat transfer at the contact surface between tissue and electrode. The proposed Dielectric Ultra-focused Oscillatory (DUO) blade is based on dielectric heating, enabling direct heating of tissue moisture without generating heat on the electrode itself. This approach inherently limits the temperature to 100 degrees C due to the principles of water vaporization. This reduces thermal damage in tissue by more than 50% compared to conventional devices and offers the advantage of preventing physical damage to implanted devices within the patient's body, such as pacemakers, leads, and batteries. The advantages of the proposed DUO blade technology were validated through in-vivo and ex-vivo tests, demonstrating its superior performance compared to existing products.
In recent years, lung cancer is a serious disease worldwide. To overcome the disease, early detection and early treatment of lung cancer are important tasks. In Japan, CT (Computed Tomography) examination is performed in many medical facilities for visual screening. However, there is a problem that huge number of images taken by CT is a burden to the doctor. Therefore, the CAD (Computer Aided Diagnosis) system is in the spotlight to reduce the burden. On the other hand, in CT screening, LDCT (Low Dose Computed Tomography) is desirable considering radiation exposure. However, the LDCT image is characterized by lower image quality at lower dose. Therefore, a CAD that can be applied to LDCT is needed. In this paper, we propose a lung field detection method and a three-dimensional (3D) registration method to generate temporal subtraction images that can be applied to LDCT images. Our method consists of the following steps: detection of lung regions using an active contour model, correction of corresponding slices between images, global matching based on the center of gravity, setting the VOI (volume of interest), and local matching on the VOI. In this paper, we apply our method to LDCT images of 7 cases. In addition, we conducted a comparison experiment with other registration methods. In the proposed method, a decrease of 7.55% in FWHM and 34.4% in sum of histogram of temporal subtraction images compared to the other method, and its usefulness was verified.
Different loading modalities can significantly affect human gait, posture, and lower limb biomechanics. This study investigated the intensity of muscle activity in the soleus muscle of the lower limbs among young healthy adult males in a unilateral loading setting on an inclined surface. Thirteen subjects carried dumbbells while walking at a fixed speed on slopes respectively. Electromyographic (EMG) changes in the bilateral soleus muscles were recorded during these experiments. One-way analysis of variance (ANOVA) and multivariate analysis of variance (MANOVA) were utilized to examine the relationships between load weight, slope angle, and muscle activity intensity. The results of this research contribute to the advancement of the field of lower limb assistive exoskeletons by addressing the data gaps related to loading on inclined surfaces, providing essential support for future assistance systems, and facilitating the development of relevant datasets aimed at enhancing terrain recognition capabilities as well as improving the mobility of device users.
Recent developments in single-cell RNA sequencing (scRNA-seq) technology have made it possible to conduct a thorough analysis of the ways in which cells express their transcriptomes in response to various physiological states and cues from the outside world. "Dropout," the phenomenon in which a gene is observed at low or moderate levels in one type of cell but not in another, poses a major analytical challenge when analyzing scRNA-seq data. Taking on the problem of dropout events in gene expression datasets is the main goal of this work. Our proposed approach consists of two main steps: first, we apply the consensus clustering-based algorithm, ccImpute, to impute dropout events, and then we utilize scCLUE, an ensemble feature selection and network similarity measurement-based single-cell clustering method, to cluster single-cell types. Our experimental results with four publicly available datasets show that the proposed approach outperforms the original scCLUE and other clustering algorithms. By addressing the problems caused by dropout events in scRNA-seq data, it improves clustering performance and facilitates more effective separation of similar cell populations.
Objective: Non-small cell lung cancer (NSCLC), as the most common type of lung cancer, has a high rate of metastasis and mortality. A large number of studies have proved that traditional Chinese medicine has the effect of anti-drug resistance and targeting of cancer cells. Platycodon platycodon can treat lung cancer, but its potential drug targets and molecular mechanisms in the treatment of NSCLC have not been fully clarified. Methods: The key components, potential core targets and main pathways of Platycodon grandiflorum were obtained by constructing the drug ingredient-disease-target network. Finally, the interaction between the selected compounds and proteins was docking and evaluated by means of molecular docking. GEPIA, HPA and Kaplan-Meier databases were used to analyze the expression of differential genes and core proteins of the core targets in tumor tissue and normal lung tissue and their survival analysis. Results: A total of 7 active ingredients and 240 component target genes were screened. Molecular docking results showed that the screened compounds had stable binding force with core targets. Differential gene analysis showed that mRNA expressions of key genes AKT1, TNF and ESR1 were up-regulated in lung cancer tissues compared with normal lung tissues. The mRNA expression of TP53 was down-regulated, and the mRNA expression of EGFR was not significantly changed. Survival analysis showed that the expression of five key genes, AKT1, TP53, TNF, EGFR and ESR1, were significantly correlated with poor prognosis (Log-rank p < 0.05). Conclusion: Platycodon grandiflorum in the treatment of NSCLC mainly involves acacetin, luteolin, Spinasterol and other key components. These compounds appear to exert their therapeutic effects by modulating the AGE-RAGE pathway, subsequently affecting proteins such as AKT1, TP53, TNF, EGFR, and ESR1, thereby contributing to the treatment of NSCLC. EGFR gene and Spinasterol compound should be considered in this study.
Down syndrome, induced by trisomy 21 (T21), manifests as cognitive impairment and significant fetal congenital anomalies. Noninvasive prenatal testing (NIPT) for trisomy 21 involves the identification of cell-free fetal DNA (cff-DNA) in maternal plasma. Current noninvasive prenatal testing (NIPT) technologies, primarily relying on next-generation sequencing (NGS), suffer from high costs and slow processing speeds. Recently, droplet digital PCR (ddPCR) stands out for its cost-effectiveness, high sensitivity, and rapid processing speed, emerging as a promising alternative to NGS for NIPT applications. But, the multiplex detection of cff-DNA from trisomy 21 using ddPCR often lacks sufficient specificity, mainly due to the complicated primers design. In this study, we present an optimized strategy for universal probe design through the screening of specific fragments at the whole-genome level. With this strategy, we designed 20 pairs of primers and a universal probe, specifically for targeting cff-DNA on chromosome 1 and 21. Our experimental results performed on simulation samples show that our developed ddPCR measurement method can effectively improve the detection specificity for cff-DNA from T21.
Gene therapy holds significant potential for treatment of different kinds of hereditary and acquired diseases. However, risk of immune response, concerns of viral vectors, including toxicity, inflammatory responses, and issues with genetic mutation cause great concerns, inhibiting this technique from taking to practical applications. To solve these problems, transfection vehicles are introduced, offering several advantages including inexpensive materials: facile synthesis process, and high transfection efficiency with low cytotoxicity. Currently most studies focus on linear structure, and herein, a highly branched poly(beta-amino ester) (HPAE) is synthesized via Michael addition approach and purified with excess diethyl ether precipitation. NMR, and GPC characterized structure and composition of HPAE. GPC data revealed that S4-TMPTA-BEDA-MPA had a molecular weight (MW) of 33 kg/mol. Further, HPAE exhibited high DNA binding, over 50%. GFP images indicates that HPAEs shows a better transfection effect at various w/w ratios on HeLa cells and SW1353 cells, and cell viability of all H-LPAEs exceeded 95% at w/w ratios of 10:1 to 80:1 in SW1353 cell, which highlights its potential application in gene therapy.
Emergency departments (EDs) are challenged by overcrowding, which affects the quality of care and increases the rate of patients leaving without being seen (LWBS). The study aims to predict LWBS cases in Local Health Authority No. 3 in Naples, Italy, a distributed ED system, using machine learning (ML) models and compare its performance with single-center EDs. Data from 53,761 patients at LHA No. 3, 83,739 at Hospital H1, and 77,607 at Hospital H2 from the year 2022 were analyzed. The dataset included gender, age, triage score, mode of arrival, and time of admission. Machine learning algorithms such as Random Forest (RF), Support Vector Machine (SVM), Naive Bayes (NB), and Logistic Regression (LR) were implemented using the KNIME Analytics platform. The primary outcome was LWBS. All ML models showed high accuracy rates above 90%. The RF model showed the highest accuracy (91.2%) and F-measure (95.3%). The study also revealed patterns in patient flow, most notably a peak in arrivals between 6:00 and 12:00. A balanced age distribution was observed, in contrast to the older patient demographics in previous studies. The ML models, particularly RF, were highly effective in predicting LWBS cases in a distributed ED system. This high predictive accuracy can contribute to the efficient allocation of ED resources and improve patient satisfaction, thereby addressing the problem of overcrowding.
