Cardiothoracic surgery patients have the risk of developing surgical site infections which cause hospital readmissions, increase healthcare costs, and may lead to mortality. This work aims to tackle the problem of surgical site infections by predicting the existence of worrying alterations in wound images with a wound image analysis system based on artificial intelligence. The developed system comprises a deep learning segmentation model (MobileNet-Unet), which detects the wound region area and categorizes the wound type (chest, drain, and leg), and a machine learning classification model, which predicts the occurrence of wound alterations (random forest, support vector machine and k-nearest neighbors for chest, drain, and leg, respectively). The deep learning model segments the image and assigns the wound type. Then, the machine learning models classify the images from a group of color and textural features extracted from the output region of interest to feed one of the three wound-type classifiers that reach the final binary decision of wound alteration. The segmentation model achieved a mean Intersection over Union of 89.9% and a mean average precision of 90.1%. Separating the final classification into different classifiers was more effective than a single classifier for all the wound types. The leg wound classifier exhibited the best results with an 87.6% recall and 52.6% precision.

Sophisticated voluntary movements are essential for everyday functioning, making the study of how the brain controls muscle activity a central challenge in neuroscience. Investigating corticomuscular control through non-invasive electrophysiological recordings is particularly complex due to the intricate nature of neuronal signals. To address this challenge, we present a novel experimental methodology designed to study corticomuscular control using electroencephalography (EEG) and electromyography (EMG). Our approach integrates a serious gaming biofeedback system with a specialized experimental protocol for simultaneous EEG-EMG data acquisition, optimized for corticomuscular studies. This work introduces, for the first time, a method for assessing brain–muscle functional connectivity during the execution of a demanding motor task. By identifying neuronal sources linked to muscular activity, this methodology has the potential to advance our understanding of motor control mechanisms. These insights could contribute to improving clinical practices and fostering the development of novel brain–computer interface technologies.

The ageing of the global population, especially in developed countries, is driving significant societal changes. In Portugal, demographic data reflect a marked increase in the ageing index. Understanding gait alterations associated with ageing is essential for the early detection of mobility decline and fall risk. This study aimed to analyse gait patterns in older adults to contribute to a biomechanical ageing profile. Thirty-six community-dwelling older adults (29 female, 7 male; mean age: 74 years) participated. Gait data were collected using the Xsens full-body motion capture system, which combines inertial sensors with biomechanical modelling and sensor fusion. Spatiotemporal and kinematic parameters were analysed using descriptive statistics. Compared to younger adult norms, participants showed increased stance and double support phases, reduced swing phase, and lower gait speed, stride length, and cadence, with greater step width. Kinematic data showed reduced peak plantar flexion, knee flexion, and hip extension, but increased dorsiflexion peaks—adaptations aimed at stability. Despite a limited sample size and lack of clinical subgroups, results align with age-related gait literature. Findings support the utility of wearable systems like Xsens in capturing clinically relevant gait changes, contributing to normative biomechanical profiling and future mobility interventions.

This article presents PrevOccupAI-HAR, a new publicly available dataset designed to advance smartphone-based human activity recognition (HAR) in office environments. PrevOccupAI-HAR comprises two sub-datasets: (1) a model development dataset collected under controlled conditions, featuring 20 subjects performing nine sub-activities associated to three main activity classes (sitting, standing, and walking), and (2) a real-world dataset captured in an unconstrained office setting captured from 13 subjects carrying out their daily office work for six hours continuously. Three machine learning models—namely, k-nearest neighbors (KNN), support vector machine (SVM), and Random Forest (RF)—were trained on the model development dataset to classify the three main classes independently of sub-activity variation. The KNN, SVM, and RF models achieved accuracies of 90.94%, 92.33%, and 93.02%, respectively, on the development dataset. When deployed on the real-world dataset, the models attained mean accuracies of 69.32%, 79.43%, and 77.81%, reflecting performance degradations between 21.62% and 12.90%. Analysis of sequential predictions revealed frequent short-duration misclassifications, predominantly between sitting and standing, resulting in unstable model outputs. The findings highlight key challenges in transitioning HAR models from controlled to real-world contexts and point to future research directions involving temporal deep learning architectures or post-processing methods to enhance prediction consistency.