An electrocardiogram (ECG) is a foundational tool for diagnosing cardiovascular diseases. However, the assessment of ECG data often depends on the subjective expertise of medical professionals, leading to challenges in accurately identifying subtle and ambiguous abnormalities. This study presents an advanced model that automates the classification of five specific cardiac diseases using AI-driven ECG analysis while validating its performance. Previous approaches have leveraged AI to classify time series data by converting these signals into image representations for analysis with convolutional neural networks (CNNs). The 1D ECG segment utilizes a bidirectional Long Short-Term Memory (bi-LSTM) network supplemented by an attention mechanism to enhance temporal information extraction. In parallel, the 2D ECG segment generates image representations through Gramian Angular Field (GAF), Recurrence Plot (RP), and Markov Transition Field (MTF) techniques. The late fusion strategy aggregates the maximum probabilities from the distinct classification outputs of the 1D and 2D analyses. Our framework has been rigorously evaluated on both 3-leads and 12-leads ECG datasets from the PTBXL database, instilling confidence in the robustness of our model. The proposed model demonstrates a significant performance improvement over state-of-the-art (SOTA) techniques, achieving enhancements in classification accuracy ranging from 0.8% to 12%. The source codes are available at: https://github.com/timeseriessignal/FusionNet

Eye tracking is a physiological technique that quantifies ocular motor behavior to analyze human visual engagement. Traditional methods rely on eye trackers to measure eye movements and spatial coordinates, typically based on video analysis. Recent advancements have explored integrating AI-driven EEG analysis to enhance eye-tracking capabilities. However, most existing approaches focus solely on real-time gaze estimation, overlooking past gaze information and eye movement kinematics. In this study, we propose an AI-based framework for real-time gaze position prediction using EEG signals that incorporate previous gaze patterns. The model architecture combines a self-attention mechanism for extracting salient features from multi-channel EEG time-series data with an LSTM network to predict current gaze positions based on past information. The training process consists of two phases: pre-training for gaze movement classification and subsequent fine-tuning for gaze position regression. Our proposed framework demonstrated an average improvement of 44% over the previous state-of-the-art model. While the earlier study indicated a gaze tracking error of 55.4 mm, the proposed model achieved a significantly lower average gaze tracking error of 24.4 mm compared to the ground truth.

Diabetes mellitus is a chronic condition marked by insufficient insulin utilization or production, causing metabolic dysregulation. If not controlled, it can cause serious complications that affect major organ systems such as the cardiovascular and ocular systems. Early diagnosis is essential for timely interventions to ensure effective glycemic control and lower the risk of complications. The present study introduced a robust and comprehensive dataset derived from nail-fold capillaroscopy. This dataset, which includes imaging and some video data from 126 individuals diagnosed with type 2 diabetes mellitus (T2DM) alongside 76 healthy controls, consisted of 3283 images obtained from diabetic subjects and 3412 images from non-diabetic participants. All images were acquired at 390× high magnification with a high resolution of 640×480 pixels, including corresponding video data, ensuring a thorough and detailed dataset for our study. The dataset was organized into two analytical tracks. The first track focused on nail-fold morphology, classifying images into four types: hairpin, crossing, tortuous, and bushy. After filtering out duplicates and low-quality images, 1279 images were selected for analysis. The second track involved binary classification for the detection of diabetes, differentiating healthy individuals from people with diabetes. This utilized the complete dataset and the augmented versions created by combining multiple images into composite formats to enhance feature representation. Additionally, we evaluated several benchmark deep-learning models such as Vision Transformers (ViT) and Convolutional Neural Networks (CNNs) for morphological classification and diabetes detection tasks. This analysis illuminated current model performance, highlighted challenges, and paved the way for future research opportunities. Importantly, this dataset is critical and holds significant potential for advancing non-invasive automated diagnostic methodologies in diabetes-related nail-fold capillary research, offering a promising future for diabetes research. The dataset is available at: https://huggingface.co/datasets/HanaNguyen/Capillary-Dataset. The github: https://github.com/urgonguyen/Capillarydataset.git.

Vision transformers (ViT) perform remarkably on large datasets but often underperform compared to convolutional neural networks (CNNs) when trained from scratch on smaller datasets. This study aimed to improve the accuracy of ViT in diabetic patient classification with the nail-fold capillaries dataset. It involved preprocessing images to extract relevant features and, unlike previous methods that used a single image, it combined multiple images into a composite for simultaneous analysis of video frames. The results demonstrated a substantial 27.53% improvement, with the F1 score increasing from 32.26% to 59.79% when using the proposed model training from scratch. Additionally, the F1 score also increased by 3.93% and achieved 98.56% with the pre-trained model using our proposed method which integrated multiple nail fold capillaroscopy images and employed ViT. Detailed analysis revealed that merging the nine images into a 3×3 format instead of a panoramic format yielded optimal results.

A stethoscope is commonly used to check the heart health of infants by measuring heart movement through sound. However, diagnosis using this equipment heavily relies on the clinician’s expertise and experience, which can result in missed early detection in cases with minimal symptoms. This study aimed to propose and verify a deep learning model that can diagnose congenital heart disease early from time series data of the heart’s audio type, which could be measured with a stethoscope. The proposed architecture is a hybrid combined model of CNN and LSTM, which involves combining 2D-CNN with LSTM layers confirmed that early detection was possible with a high accuracy of 97.8%.

Velocity-based training can be used for other valuable purposes, such as providing instantaneous feedback to the athletes to promote performance improvements. Although this training tool appears to have many practical functionalities, coaches should not get distracted by the technology and forget what they are actually there to do, which is to coach. This study aimed to propose an extended relative phase (ERP) algorithm for assessing velocity-based training through human pose estimation (HPE) and verifying its feasibility while demonstrating the difference between experts and novices. The coordinating relationship between the trunk and lower limbs during the back barbell squat in this study could be illustrated as a circle. ERP had two leading indices for time series; a Relative phase angle (RPAngle) defined the collaboration between angular displacements and velocities for each joint, and a Relative phase distance (RPDistance) was added to illustrate the stability during the exercise. Through experiments of the back barbell squat with 15kg, a periodicity and amplitude for two time series of RPAngleand RPDistanceenabled us to quickly understand the difference in exercise pose between experts and novices. It was concluded that the proposed method allowed coaches to simultaneously assess the angle and angular velocity during the training through ERP despite any particular technical knowledge.

It is crucial to diagnose diabetes in the early phase because it can cause serious complications if not treated early. Nailfold capillaries are capillaries that are located under the proximal nail fold area. Studies have shown that different types of abnormal patterns of nail fold capillaries can be important factors when diagnosing various diseases or conditions. Therefore, this study proposes a system that can detect the nailfold capillary morphology of type 2 diabetes mellitus in a non-invasive way utilizing an object detection model. We trained an object detection model YOLOv5 with bi-directional feature pyramid network (Bi-FPN) as the neck part, which is an improved version of the open-source object detector YOLOv5 to detect capillaries and their corresponding morphology types including hairpin, crossing, bushy, and tortuous capillaries with mAP@50 0.74. We expect the proposed system to aid in the early diagnosis of diabetes in the future.

This paper proposes a web-based extended reality (XR) platform integrating exhibitions based on augmented reality (AR) and virtual reality (VR). The proposed platform includes an XR cloud for managing and sharing data in XR spaces, a VR client for users, and an AR client for real-world users. The XR cloud is a web-based VR service responsible for managing objects and spatial information related to real and virtual spaces while simultaneously facilitating real-time information sharing and synchronization between AR and VR users. The VR client offers a virtual exhibition environment based on the virtual space and information for VR users. The AR client overlays virtual information over objects in a real space, allowing for intuitive interaction with them. The proposed XR platform was implemented with Oculus Quest 2 for VR and HoloLens 2 for AR and was evaluated in a small exhibition environment. In this experimental environment, VR and AR users were able to participate in the same exhibition from different spaces and share their experiences in real time.

We have an interest in the handling posture at the logistic workplace. Although a lot of studies on compressive and shear forces on lumbar spine have been reported, there was little studies on the difference of asymmetric lifting between logistic workers and beginners with the comparison of measured results. The main purpose of this study was to analyze the difference between ten healthy workers (Group 1) and five beginners (Group 2) during the asymmetric lifting through measurement of center of pressure (COP) locations. We performed experiments to move the 18 kg load to the left side under the condition of closed eyes. Through analysis results of COP locations, we found that Group 1 had the difference from Group 2. Most of COPs of Group 1 was located in the right side during the trunk angle to twist to the left side. We reported all of experimental results in this study.

We have an interest in the relationship between the human postural balance and back injuries. Although back injuries are the number-one problem facing the workforce in the industry of carrying loads, there appears to be no firm idea on how this will deal with in practice. It is known that the brain plays an important role in maintenance of standing posture. In this paper, we focus on evaluating how the human postural balance is affected under three conditions of lifting loads. Especially, we evaluate the human postural balance during the rising time. We use the spectrum analysis for evaluating the postural balance by using the Nintendo Wii balance board. Through all of experiments, we found the tendency that the heavier loads are, the faster and larger the body swing for balancing the posture during the rising time is. Additionally, the case excluding bending the knee shows the larger swing than that including bending the knee. From all of results, we could hypothesize that an expert of carrying loads may make a balance for the posture during rising time as soon as possible by using the smaller swing motion. We show all of experimental results.

This paper newly proposes the four-fingered robot hand with dual turning mechanism where two and two other fingers can independently rotate inner and outer circles with the common center, respectively. Due to this mechanical configuration, it has the particular rotating axis where the manipulation around the axis can be completely decomposed into the velocity control around the axis and the internal force control in the contact plane. We achieved a manipulation task around the axis with the time of 0.8[sec] for one rotation, while it is relatively slow for another axis.