Search

This work aims to compare the performance of Machine Learning (ML) and Deep Learning (DL) algorithms in detecting users’ heartbeats on a smart bed. Targeting non-intrusive, continuous heart monitoring during sleep time, the smart bed is equipped with a 3D solid-state accelerometer. Acceleration signals are processed through an STM 32-bit microcontroller board and transmitted to a PC for recording. A photoplethysmographic sensor is simultaneously checked for ground truth reference. A dataset has been built, by acquiring measures in a real-world set-up: 10 participants were involved, resulting in 120 min of acceleration traces which were utilized to train and evaluate various Artificial Intelligence (AI) algorithms. The experimental analysis utilizes K-fold cross-validation to ensure robust model testing across different subsets of the dataset. Various ML and DL algorithms are compared, each being trained and tested using the collected data. The Random Forest algorithm exhibited the highest accuracy among all compared models. While it requires longer training time compared to some ML models such as Naïve Bayes, Linear Discrimination Analysis, and K-Nearest Neighbour Classification, it keeps substantially faster than Support Vector Machine and Deep Learning models. The Random Forest model demonstrated robust performance metrics, including recall, precision, F1-scores, macro average, weighted average, and overall accuracy well above 90%. The study highlights the better performance of the Random Forest algorithm for the specific use case, achieving superior accuracy and performance metrics in detecting user heartbeats in comparison to other ML and DL models tested. The drawback of longer training times is not too relevant in the long-term monitoring target scenario, so the Random Forest model stands out as a viable solution for real-time ballistocardiographic heartbeat detection, showcasing potential for healthcare and wellness monitoring applications.

The lack of physical exercise is among the most relevant factors in developing health issues, and strategies to incentivize active lifestyles are key to preventing these issues. The PLEINAIR project developed a framework for creating outdoor park equipment, exploiting the IoT paradigm to build “Outdoor Smart Objects” (OSO) for making physical activity more appealing and rewarding to a broad range of users, regardless of their age and fitness. This paper presents the design and implementation of a prominent demonstrator of the OSO concept, consisting of a smart, sensitive flooring, based on anti-trauma floors commonly found in kids playgrounds. The floor is equipped with pressure sensors (piezoresistors) and visual feedback (LED-strips), to offer an enhanced, interactive and personalized user experience. OSOs exploit distributed intelligence and are connected to the Cloud infrastructure by using a MQTT protocol; apps have then been developed for interacting with the PLEINAIR system. Although simple in its general concept, several challenges must be faced, related to the application range (which called for high pressure sensitivity) and the scalability of the approach (requiring to implement a hierarchical system architecture). Some prototypes were fabricated and tested in a public environment, providing positive feedback to both the technical design and the concept validation.

Artificial Intelligence in combination with the Internet of Medical Things enables remote healthcare services through networks of environmental and/or personal sensors. We present a remote healthcare service system which collects real-life data through an environmental sensor package, including binary motion, contact, pressure, and proximity sensors, installed at households of elderly people. Its aim is to keep the caregivers informed of subjects’ health-status progressive trajectory, and alert them of health-related anomalies to enable objective on-demand healthcare service delivery at scale. The system was deployed in 19 households inhabited by an elderly person with post-stroke condition in the Emilia–Romagna region in Italy, with maximal and median observation durations of 98 and 55 weeks. Among these households, 17 were multi-occupancy residences, while the other 2 housed elderly patients living alone. Subjects’ daily behavioral diaries were extracted and registered from raw sensor signals, using rule-based data pre-processing and unsupervised algorithms. Personal behavioral habits were identified and compared to typical patterns reported in behavioral science, as a quality-of-life indicator. We consider the activity patterns extracted across all users as a dictionary, and represent each patient’s behavior as a ‘Bag of Words’, based on which patients can be categorized into sub-groups for precision cohort treatment. Longitudinal trends of the behavioral progressive trajectory and sudden abnormalities of a patient were detected and reported to care providers. Due to the sparse sensor setting and the multi-occupancy living condition, the sleep profile was used as the main indicator in our system. Experimental results demonstrate the ability to report on subjects’ daily activity pattern in terms of sleep, outing, visiting, and health-status trajectories, as well as predicting/detecting 75% hospitalization sessions up to 11 days in advance. 65% of the alerts were confirmed to be semantically meaningful by the users. Furthermore, reduced social interaction (outing and visiting), and lower sleep quality could be observed during the COVID-19 lockdown period across the cohort.

This work presents a methodology to analyze and segment both seismocardiogram (SCG) and ballistocardiogram (BCG) signals in a unified fashion. An unsupervised approach is followed to extract a template of SCG/BCG heartbeats, which is then used to fine-tune temporal waveform annotation. Rigorous performance assessment is conducted in terms of sensitivity, precision, Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) of annotation. The methodology is tested on four independent datasets, covering different measurement setups and time resolutions. A wide application range is therefore explored, which better characterizes the robustness and generality of the method with respect to a single dataset. Overall, sensitivity and precision scores are uniform across all datasets ( p > 0.05 from the Kruskal–Wallis test): the average sensitivity among datasets is 98.7%, with 98.2% precision. On the other hand, a slight yet significant difference in RMSE and MAE scores was found ( p < 0.01 ) in favor of datasets with higher sampling frequency. The best RMSE scores for SCG and BCG are 4.5 and 4.8 ms, respectively; similarly, the best MAE scores are 3.3 and 3.6 ms. The results were compared to relevant recent literature and are found to improve both detection performance and temporal annotation errors.

This paper presents an unsupervised methodology to analyze SeismoCardioGram (SCG) signals. Starting from raw accelerometric data, heartbeat complexes are extracted and annotated, using a two-step procedure. An unsupervised calibration procedure is added to better adapt to different user patterns. Results show that the performance scores achieved by the proposed methodology improve over related literature: on average, 98.5% sensitivity and 98.6% precision are achieved in beat detection, whereas RMS (Root Mean Square) error in heartbeat interval estimation is as low as 4.6 ms. This allows SCG heartbeat complexes to be reliably extracted. Then, the morphological information of such waveforms is further processed by means of a modular Convolutional Variational AutoEncoder network, aiming at extracting compressed, meaningful representation. After unsupervised training, the VAE network is able to recognize different signal morphologies, associating each user to its specific patterns with high accuracy, as indicated by specific performance metrics (including adjusted random and mutual information score, completeness, and homogeneity). Finally, a Linear Model is used to interpret the results of clustering in the learned latent space, highlighting the impact of different VAE architectural parameters (i.e., number of stacked convolutional units and dimension of latent space).

This paper introduces technical solutions devised to support the Deployment Site - Regione Emilia Romagna (DS-RER) of the ACTIVAGE project. The ACTIVAGE project aims at promoting IoT (Internet of Things)-based solutions for Active and Healthy ageing. DS-RER focuses on improving continuity of care for older adults (65+) suffering from aftereffects of a stroke event. A Wireless Sensor Kit based on Wi-Fi connectivity was suitably engineered and realized to monitor behavioral aspects, possibly relevant to health and wellbeing assessment. This includes bed/rests patterns, toilet usage, room presence and many others. Besides hardware design and validation, cloud-based analytics services are introduced, suitable for automatic extraction of relevant information (trends and anomalies) from raw sensor data streams. The approach is general and applicable to a wider range of use cases; however, for readability’s sake, two simple cases are analyzed, related to bed and toilet usage patterns. In particular, a regression framework is introduced, suitable for detecting trends (long and short-term) and labeling anomalies. A methodology for assessing multi-modal daily behavioral profiles is introduced, based on unsupervised clustering techniques. The proposed framework has been successfully deployed at several real-users’ homes, allowing for its functional validation. Clinical effectiveness will be assessed instead through a Randomized Control Trial study, currently being carried out.

Environmental sensors are exploited in smart homes for many purposes. Sensor data inherently carries behavioral information, possibly useful to infer wellness and health-related insights in an indirect fashion. In order to exploit such features, however, powerful analytics are needed to convert raw sensor output into meaningful and accessible knowledge. In this paper, a complete monitoring architecture is presented, including home sensors and cloud-based back-end services. Unsupervised techniques for behavioral data analysis are presented, including: (i) regression and outlier detection models (also used as feature extractors for more complex models); (ii) statistical hypothesis testing frameworks for detecting changes in sensor-detected activities; and (iii) a clustering process, leveraging deep learning techniques, for extracting complex, multivariate patterns from daily sensor data. Such methods are discussed and evaluated on real-life data, collected within several EU-funded projects. Overall, the presented methods may prove very useful to build effective monitoring services, suitable for practical exploitation in caregiving activities, complementing conventional telemedicine techniques.

A balanced diet is the key to a healthy lifestyle and is crucial for preventing or dealing with many chronic diseases such as diabetes and obesity. Therefore, monitoring diet can be an effective way of improving people’s health. However, manual reporting of food intake has been shown to be inaccurate and often impractical. This paper presents a new approach to food intake quantity estimation using image-based modeling. The modeling method consists of three steps: firstly, a short video of the food is taken by the user’s smartphone. From such a video, six frames are selected based on the pictures’ viewpoints as determined by the smartphone’s orientation sensors. Secondly, the user marks one of the frames to seed an interactive segmentation algorithm. Segmentation is based on a Gaussian Mixture Model alongside the graph-cut algorithm. Finally, a customized image-based modeling algorithm generates a point-cloud to model the food. At the same time, a stochastic object-detection method locates a checkerboard used as size/ground reference. The modeling algorithm is optimized such that the use of six input images still results in an acceptable computation cost. In our evaluation procedure, we achieved an average accuracy of 92 % on a test set that includes images of different kinds of pasta and bread, with an average processing time of about 23 s.