International Conference on Biomedical and Health Informatics 2022

Type 2 diabetes mellitus (T2DM) is a disease that affects more than 380 million people worldwide. In this study, we developed a model, for these type of patients, that predicts blood glucose values over long prediction horizons (PHs), whose existence in the literature is almost nonexistent. These horizons allow patients to be warned in advance so that they can take action to avoid dangerous health situations. We used data from 3 of the 10 real patients available to test the implemented models. The overall results for the best model (simple Recurrent Neural Network) were: 34.82 mg/dL for root mean square error (RMSE) and 18.33% for mean absolute percentage error (MAPE) (PH = 2h); 46.59 mg/dL for RMSE and 24.35% for MAPE (PH = 4h).

Medical image segmentation plays a crucial role in diagnosis assistance. In previous works, we proposed a classification method called Type-2 Label-based Fuzzy Predicate Classification (T2-LFPC), which generates Interval-Valued Membership Functions (IVMF) and fuzzy predicates. They can be analyzed to interpret the images. In this work, a methodology is proposed to study the semantic of IVMF generated from brain MRI as input of the T2-LFPC. It is possible to understand both membership functions and predicates by visual inspecting positions and shapes of the IVMF. Some changes are applied on the images. Transformations include: zero mean additive noise, contrast-stretching and brightness increase and decrease. Changes in the images by transformations are reflected in the histograms of the pixels belonging to white matter, gray matter, and cerebrospinal fluid, in the IVMF and the values of their measures. Therefore, as changes are reflected in the IVMF as expected, the methodology proposed here could be considered suitable for image analysis.

Lung cancer is the leading cause of cancer death in the world. Accurate determination of the EGFR (epidermal growth factor receptor) mutation status is highly relevant for the proper treatment of this patients. Purpose: The aim of this study was to predict the mutational status of the EGFR in non-small cell lung cancer patients using radiomics features extracted from PET-CT images. Methods: Retrospective study that involve 34 patients with lung cancer confirmed by histology and EGFR status mutation assessment. A total of 2.205 radiomics features were extracted from manual segmentation of the PET-CT images using pyradiomics library. Both computed tomography and positron emission tomography images were used. All images were acquired with intravenous iodinated contrast and F18-FDG. Preprocessing includes resampling, normalization, and discretization of the pixel intensity. Three methods were used for the feature selection process: backward selection (set 1), forward selection (set 2), and feature importance analysis of random forest model (set 3). Nine machine learning methods were used for radiomics model building. Results: 35.2% of patients had EGFR mutation, without significant differences in age, gender, tumor size and SUVmax. After the feature selection process 6, 7 and 17 radiomics features were selected, respectively in each group. The best performances were obtained by Ridge Regression in set 1: AUC of 0.826 (95% CI, 0.811–0.839), Random Forest in set 2: AUC of 0.823 (95% CI, 0.808–0.838) and Neural Network in set 3: AUC of 0.821 (95% CI, 0.808–0.835). Conclusion: The radiomics features analysis has the potential of predicting clinically relevant mutations in lung cancer patients through a non-invasive methodology.

Cancer is a major threat to humankind and a leading cause of mortality worldwide. According to World Health Organization estimates in 2020, breast cancer is a top-tier cancer and a substantial cause of death in women. The early identification of breast cancer can effectively reduce risk factors and mortality. Recent studies in risk analysis were focused on the stand-alone impact of genomic data ignoring the influence of environmental factors. In our proposed method, we studied the breast cancer risk assessment using multi-omics with epigenetic factors and a deep learning model. Our model outperformed existing breast cancer detection methods and stage identification on data collected from TCGA-GDC datahub. However, the vitality analysis model could not produce significant results mainly due to non-availability of sufficient quality data on survival information. The proposed model validates the crucial role of DNA methylation in pre-symptomatic breast cancer risk analysis.

This paper presents a novel example of depression prediction, merging cognitive science with data-driven machine learning. Behavioral economic features were engineered from a short picture rating task. Relative Preference Theory was applied to rating data for quantifying the degree to which participants liked, disliked, or were neutral to several types of pictures; thus, behavioral economic variables including loss aversion, risk aversion, and 13 others that are amenable to psychological interpretation were mined. These variables were features of a logistic regression predictive model that targeted depression in a population-based sample (N = 281) with high test accuracy and no overfitting. Per our review of the literature, we cannot identify other papers that explicitly use behavioral economic features to predict depression with machine learning.

Epileptic seizures need to be predicted with sufficient time to allow the patients to prevent clinical symptoms by taking their prescribed anti-epileptic medications or via neurostimulation. In this context, automatic seizure prediction techniques with high sensitivity and specificity are needed. In this work, we present a machine learning method to predict epileptic seizures by analyzing the pre-ictal stage of scalp EEG recordings. The TUH EEG Seizure Corpus files were used to train and test the technique. After filtering and channel normalization, statistical parameters from the Teager energy operator, the absolute band power, the Hjorth parameters, and the kurtosis of the signal were extracted from each channel. Four machine learning classifiers were implemented after a feature selection step based on inter-feature cross-correlation values: Random Forest, XGBoost, Support Vector Machine, and K-nearest neighbors. The XGBoost model yielded the best performance metrics with the validation set: 99.84% in accuracy, 100% in precision, 99.6% in recall, 100% in specificity, and 0.998 in the area under the receiving operating characteristics curve (AUC). Ultimately, these methods can detect subtle changes in scalp EEG records that forecast an epileptic seizure.

Depression is a globally known disease with a great impact on the suicide rate. However, this can be an early diagnostic by observing the behavior of the patients through the time. In this research, we studied the linguistics and visual features of depressive mood during COVID-19 pre and post-pandemic based on Flickr posts. We implemented the significant advances in text-based sentiment analysis and image classification using Natural Language Processing (NLP), histograms and deep learning strategies to characterize some of the main patterns of depression. We demonstrate that user’s behavior in social media had a relevant impact during pandemics, since the main patterns change drastically between periods. For images, we found that in pre-pandemic, user posts were more uniform in color distribution and with medium to low levels of light intensity. Besides, the scenes were more outside activities like. For text, we found that the topics and general sentiment were always depressive and with negative connotation, however, during pre-pandemic they described attributes of the symptomatology of depression pathology, while in post-pandemic are more related to the product of isolation and fear.

Atrial Fibrillation (AF) is the most dangerous arrhythmia for human health. The beating in the top two chambers is irregular (asynchronous and faster) because electrical impulses suddenly start firing in the atria, overriding the natural pacemaker of the heart. Consequently, patients with AF may require periodic check-ups that involve performing a standardized twelve-lead Electrocardiogram (ECG) exam. Therefore, to assist doctors with AF detection, conventional algorithms have been developed based on the irregularity of the R-R segment or detecting the P-wave absence. In addition, Convolutional Neural Networks (CNN) have been employed to detect the AF, showing promising results. However, the training of these networks requires large amounts of ECG data. Furthermore, the careful design and tuning of a deep neural network model consumes a considerable amount of time and computational resources. In this work, we employ the Inductive Transfer Learning (ITL) method and a ResNet18. This network is pre-trained with the public icentia11k database in a beat classification job. The fine-tuning stage was performed by merging the PhysioNet data challenges (2020, 2021) to configure a single dataset. We obtain an accuracy greater than 89% using a single II-lead. We conclude that by employing the ITL technique, a ResNet18, and a single II-lead is possible to classify the AF and AFL arrhythmias. Therefore, this work provides significant insight into the ITL and the reliability of employing a single lead to classify the AF. This technique may be useful in real-time AF classification when employing a single ECG lead in a mobile device.

The performances achieved by machine learning models have demonstrated a high potential to revolutionise the support of clinical decision-making. However, in opposition to the high performance, the lack of transparency of these models has been pointed out as one of the major limits to their adoption in daily healthcare applications. If accomplished, transparency issues, including interpretability and explainability would contribute to a better understanding of how a model works, providing a justification for its outcomes, increasing confidence in the use of such models, and effectively assisting clinicians in decision-making.Explainable Artificial Intelligence, as a recent research field, has proposed several approaches aiming at creating more explainable models, whilst maintaining high performances. This work presents a short overview of the state-of-the-art, as well as the current challenges associated with the interpretability and explainability of machine learning models. Further, future directions for interpretable machine learning in the clinical domain are outlined, in particular the introduction of reliability measures to increase the confidence of professionals, and the development of hybrid solutions, able to integrate á prior domain knowledge (clinical evidence) in the data-driven process.

Unsustainable health costs impose a new health care paradigm, where the support to clinical decision making assumes a critical importance. In this context, several machine learning risk assessment models have been developed in order to support a proper patients’ stratification. Although their superior performances, machine learning-based risk assessment models have faced strong difficulties to obtain the trust of professionals in their application in daily clinical practice. This work proposes a strategy able to address some of the major limitations of such models: i) interpretability; ii) personalization; iii) ability to incorporate new knowledge/new risk factors.An hybrid scheme is developed, combining knowledge-driven methods (to create an interpretable set of rules for the general population) with data-driven methods (to select the most suitable subset of rules for each individual). Three main steps can be identified: i) derivation of an initial set of rules directly from current clinical evidence and/or data, ii) personalized scheme where a subset of the initial rules is identified as the most adequate one to classify a given patient; iii) an ensemble voting strategy based on the outputs of the previously selected rules. Moreover, the strategy demonstrates a high flexibility to incorporate new risk factors (in this case the inflammation biomarker), through the definition of additional rules.This strategy was applied in the context of cardiovascular disease, namely on the risk stratification of Acute Coronary Syndromes patients. It was validated based on a real dataset composed of N = 1544 patients, admitted in the Cardiology Unit of Coimbra Hospital and Universitary Centre, achieving a SE = 0.763 and SP = 0.778.

Actionable information extraction has recently become a very attractive research area. Information extraction has been around for a while, but usually the actions that can be triggered or supported by the extracted information have been seldom considered. Currently, a plethora of algorithms is used to create models that provide information extraction abilities from different types of data with different types of applications. In this paper we propose to use a distillation method based on decision-trees that transfers knowledge from black-box models to more interpretable models to understand the decision patterns in different applications. Prediction results on a credit score problem show that it is possible to use white-box methods that work on black-box results to show the potential interpretation of the decision patterns.

The issue of healthy aging and the constant increase in the number of older adults is of concern from different points of view. The work arose from the need to analyze the state of the art of tests for the evaluation and cognitive training for older adults at a distance or in virtual mode, knowing which were the main digital applications, their different modes of operation, and their psychometric characteristics, to analyze and project its application in Chile. From this, it was decided to develop a pilot application, which included five items or tests: the Trail Making Test (A and B), the Bell Test, the Visual Categorization task, and the Anxiety (Hamilton) and Depression (Yesavage). The application was developed on the Psychopy platform (Python), it is operative in a web service, and a first functionality, usability and user experience test was carried out with five adults. Statistical and comparative tables were prepared based on the results.

Several corpora and multimedia databases have been developed for research purposes. Some involve numbers, emotional speeches, or recordings of spontaneous or acted conversations with volunteers or professional speakers. Other speech databases have been created with children with normal speech or speech with language disorders for rehabilitation purposes or linguistic studies. However, it is not common to find databases in Spanish that contain vowels pronounced by children or young people. On the other hand, information about speakers such as their age or height is also difficult to find. This information is relevant in studies where the correlation between anthropometric data and other types of variables is needed. To meet this need, we created VocES, a database of child and youth vowels in Spanish for research purposes. This article describes the technical aspects of the obtained database and relates some research applications. VocES contains 1,665 recordings of the five vowels of Spanish from subjects aged 3 to 18, and also includes information on the age, height, and gender of the speakers.

According to current estimates, patients with cognitive disorders (for example due to stroke, Parkinson’s and Alzheimer’s disease), dismissed from hospitalization, are subject to a pathological recurrence due to the absence of post-discharge rehabilitation activities; they lose the functional recovery obtained during hospitalization and neurological and cognitive functions often decline. However, telerehabilitation may support effective post-hospitalization treatments by allowing physicians to assign tasks remotely to patients and to monitor their progress, thus reducing overall costs both for national healthcare systems and for patients. Nevertheless, remote and at-home rehabilitation pose several challenges, especially, with the monitoring of the level of engagement of patients during rehabilitation therapy execution. Indeed, while remote rehabilitation has several advantages over standard clinical routine, it is necessary to strictly follow rehabilitation exercises to prevent them from being ineffective.Given these premises, the REHASTART project proposes a platform based on a deep learning framework for monitoring patients’ engagement through automated classification of facial expressions and attention levels during exercise execution. More specifically, the proposed approach foresees reliable gaze estimation and emotion recognition through vision transformers. Performance analysis shows that the proposed approach achieves satisfactory accuracy in both facial expression classification and gaze estimation when tested on patients showing motion and cognitive deficits. The results of the deep learning model may be used as a feedback to physicians to monitor training sessions, and to tune them suitably to maximize the effectiveness for each patient.

Alzheimer’s Disease and Related Dementias (ADRD) create unimaginable stress and financial burden in the lives of informal caregivers, including family and friends. However, the current healthcare system is not designed to incorporate caregivers in the healthcare team model for an early diagnostic process. The objectives of the focus group study are twofold: 1) to understand the digital information needs of caregivers to develop a user-friendly digital health app for the early diagnosis of ADRD patients, and 2) to assess the willingness of caregivers to be included in the early diagnosis process using a digital health app. We conducted a focus group study to understand caregivers’ digital information needs and willingness to participate in the early diagnosis of ADRD. A total of 25 caregivers participated in the five focus group sessions. Data analysis revealed three main themes: 1) eagerness to be involved with the healthcare team for diagnosis, 2) facilitating digital communication for early diagnosis, and 3) chronicling the journey. Our unique finding demonstrates caregivers’ willingness to participate in diagnostics by utilizing innovative digital health applications. The result of this study informs an intuitive digital interface based on user-centered design.

The National Deaths Informatics System (SINADEF) of Peru was born as a response to the illicit practices of false health professionals, the registration of deceased as living persons, illicit collection of benefits, lack of verification of the authenticity of a death certificate, unvalidated data of the deceased, and the possibility of not record the death of a citizen. According to information from the year 2022, this system is implemented in 100% of the regions of Peru. Despite this, there are relevant problems in SINADEF, such as the lack of internet connectivity in areas far from cities, the need to use computers and printers, and technical problems together with the maintenance of the application in its operation. The present research focused on analyzing the usability perceived by the SINADEF users through a questionnaire based on the Questionnaire for User Interaction Satisfaction (QUIS) methodology. Demographic characteristics were collected from a total of 271 participants from different regions of Peru, with 6 main sections for the usability assessment. The results showed that the GeneralSystem Usability Score of SINADEF is about 6.64 points out of 9, thus concluding that the work was adequate and could be adapted to the characteristics of the National Deaths Informatics System, as well as to the health professionals who participated in the research by different levels of complexity on their healthcare establishments.

This paper reports on the implementation of a continuous monitoring system using a non-invasive integrated sensor-embedded sheet device. The system will have the enablement of the device itself, the implementation of a web platform for the real-time management and visualization of a patient/user and of the realization of a pilot test on real patients to evaluate the bed sheet in question. The overall objective of this study is to improve palliative care of patients through non-invasive and remote sensor monitoring by incorporating a bed sheet in the patient’s bed. This study was done in collaboration with the company Healthtracker Analytics. This was achieved by processing the Scientific Ethics Committee (CEC) documentation provided by RedSalud with the objective of obtaining their approval and subsequently the patient’s signed informed consent in order to proceed with the implementation of the device in the individual’s home in the pilot test.

Hospital facilities today have to deal with exponentially increasing volumes of batch and streaming data, comprising a variety of structured, unstructured, and semi-structured data types, originating from an ever-increasing number of sources and disparate devices located in healthcare facilities, such as intensive care units.At the same time, doctors demand faster and easier access to reliable and up-to-date data to make accurate decisions.Finding the right data sets and making them available for analysis is often a complicated process that further slows down clinical decisions. That is compounded by regulatory compliance and security controls that must be manually applied at every step of the data lifecycle, from data generation to analytics.Smart UPC is an interoperable solution that has emerged as an architecture for managing data with machine learning (ML) to overcome these challenges. The layered architecture focuses on making data readily available in a dashboard for Intensive Care Unit users, improving insights based on the use of AI and leveraging automation to simplify administration, the safety and quality of health care, complying with cybersecurity requirements and under very strict ethical standards.Smart UPC, based on a proprietary HL7 FHIR platform offers a unified, scalable, reliable and comprehensive view of ICU clinical data. It is a solution to support applications in decision making in a complex environment, with high critical demand.

In this work, we propose a method for the automatic review of Chilean medical accounts. The medical bills are governed by the Free Choice Modality and Tariff Regulations (by its Spanish acronym MLE). We present a methodology for structuring the regulations so that the algorithm can validate compliance with the regulations in a medical account. The rules are structured using an Excel file containing information such as the code, description, type, and group of each rule. A subset of the Regulations was selected, which was structured with the proposed methodology. To test the compliance of a medical account in PDF format, an algorithm written in Python reads the file and identifies the information of each service. Compliance with these regulations was evaluated in a set of simulated medical accounts. The results showed the proper functioning of the algorithm, in much less time than the current method of reviewing medical accounts. Furthermore, the methodology enables an easy update of the MLE versions.

The objective of this work is to develop and implement a chatbot aimed at the informal caregiver of patients in GES 4 (Explicit Health Guarantees), to be used by health centers, companies, or institutions that provide support to caregivers. The chatbot uses natural language processing (NLP) tools, particularly word-embeddings, to process user queries. Among all the options for developing the chatbot, SBERT was selected as be the best option for performing sentence embeddings. These embeddings are used to make the machine process the incoming queries, going from a natural language to one understood by the computer. Then, in conjunction with a palliative care expert, a database of frequently asked questions and their answers was created and integrated into the algorithm. The chatbot was evaluated by an expert in the field five times to check the correct functioning of the chatbot for questions written correctly, with spelling errors and using non-existent words. In all cases, consistent answers to the questions asked by the user were obtained. The chatbot allows continuous and direct communication 24 h a day between the health center and the patient.

In this work we present the design and early development of a portable Electrical Impedance Tomography (EIT) system consisting of a small stand-alone device capable of generate a stimulus and take measurements intended for EIT image reconstruction. The system follows the typical EIT architecture of stimulate the tissue with a known alternate current and then measure voltages in the boundary. All these stages have been implemented using only integrated circuits and a microcontroller. The current source generates a constant 1mApp sinusoidal current at 50 kHz and the ADC can measure differential voltages at 3 MS/s in multiple channels. The system has 16 current outputs/voltage inputs that connects with electrodes that would attach to the body. This design is intended for noninvasive breathing monitoring and ambulatory pulmonary function test.

Heart Rate Variability (HRV) is normally detected and extracted from the Electrocardiogram (ECG) and is a significant index of a human body status. The ECG normally use invasive skin contact sensors. A new non-invasive approach of extracting HRV based on wavelet transform from heart movement acquisition method using a 24 GHz radar sensor is presented. An ECG and radar heart movement joint database from 30 subjects is used in this study to validate the performance of the extraction algorithm. It is shown that the proposed radar recording processing method has a consistent performance in locating heart rate, showing the direct link between radar beats and associated ECG dataset R peaks.

Two designs of four-layered non-contact capacitive electrodes were used to perform biomedical measurements on a physical prototype. For this purpose, a low-power wireless biomedical sensor system was used. The goal of this research was to compare these measurements with the results obtained from a stationary and low-frequency analysis of the capacitive coupling between the aforementioned electrodes and a concentric cylinder model of the upper arm, implemented in the CST Studio Suite® software. The results of the capacitive coupling dependence on the fabric thickness and its dielectric permittivity, the distance between the electrode and fabric, as well as detection surface radius, have confirmed the applicability of such finite element method simulations for qualitative system analysis without the necessity of creating a physical prototype, as well as provided guidelines on the further development of non-contact capacitive electrodes.

The rise of cycling as a common form of daily exercise combined with recent advances in wearable systems provides an opportunity for advanced activity monitoring using custom-tailored wireless biomonitoring systems. In this study, a low-power, small-size, lightweight battery-powered wireless bio-medical sensor system using non-contact capacitive electrodes is proposed. Two main and several secondary case studies were evaluated using three test scenarios. The best results were obtained using a specially crafted cycling glove with an embedded body bias graphite electrode.

Development in the field of smart wearable products for monitoring daily life health status is beginning to spread in society. Textile electronic methods are improving and facilitating the manufacturing of sensorized garments. This paper evaluates a newly developed t-shirt incorporating electronic sensing and interconnecting elements integrated into the T-shirt with textile-friendly techniques sensorized with a Movesense device for monitoring ECG and HR and activity. The measurement results obtained from the t-shirt are entirely in agreement with the measurements obtained with other textile garments and encourage us for a near future where wearable sensors are just textile garments sensorized seamlessly without suboptimal textile-electronic integrated elements.

Digital forensics is the branch of forensic sciences that deals with the analysis and examination of evidence obtained from digital devices. With the advent of “healthcare 4.0” and medical IoT wearables and devices, digital forensics will increasingly have to deal with medical equipment and devices. These new sources of digital evidence pose new challenges to the forensic examiners: a wide range of devices, from simple wearables that gather health-related signals, to implanted devices, robots that perform surgery, and complex imaging machinery, they are all potential sources of digital evidence. The data structures used by these systems to store information, and the protocols that they use to communicate are not always documented, occasionally forcing an examiner to reverse engineer the meaning behind the raw bytes found in the storage media or network dumps. Finally, devices used in healthcare use a variety of operating systems, some common but customized by the manufacturer, and some nice and relatively unknown. The systems software can potentially be and older, unmaintained version, which was certified in the past, and cannot be updated. To understand the data and information extracted from these devices and equipment, the digital forensics expert will need the aid of medical experts and could also potentially require help from engineers and technicians that know the inner workings, mechanisms, and physical, chemical, and biological phenomena that come into play in their operation and use. It is thus necessary to work on updating and adapting the existing guidelines for digital forensics analysts and incident responders to consider the specific issues they will encounter when working on medical devices and equipment. In this work we propose a starting point and considerations that can help lay the groundwork needed and lay the path forward for future work.

In Peru, nearly 16% of people over 20 years old suffer from a cardiovascular disease (CVD), which represents the third cause of mortality around the country. The National Statistic and Informatic Institute (INEI) demonstrates that the ratio between the number of patients per doctor is insufficient with 16.8 doctors per 10 thousand inhabitants, whereas it should be near to 23 doctors per 10 thousand inhabitants. A similar situation is replicated in Pontifical Catholic University of Peru (PUCP) medical center, with a ratio of one cardiologist per 37,000 members of the university community. For this reason, it is designed a health-telemonitoring implementation system for electrocardiograph wearable medical devices, based in life-cycle management, in order to prevent pathologies related to CVD, such as critical ischemic symptoms and heart failure. PulseBit Ex (PEX) was used as the wearable electrocardiograph commercial device, while PUCP community and medical center were a reference for the definition of clinical requirements. The results include protocols for performance, electrical safety and usability tests for validation and maintenance stages; the operation and support flowchart consisting of five phases: admission, induction, monitoring, patient report and treatment follow-up; training, preventive and corrective maintenance, and device end of life protocols as well as a website prototype which its function is to connect the device’s interface with the health personnel. Feedback obtained from PUCP Health Center during the validation of the system implementation methodology, allowed to obtain results that suggest the scalability of the project to other Peruvian health centers. In this sense, the present work achieves the objectives initially set and leaves a precedent for possible future research, in local and international context.

A far too frequent practical challenge in clinical informatics research and method development for acquiring vital signs is the extraction and synchronisation of signals from proprietary devices for the clinical monitoring of patients. In an ongoing study evaluating methods for video-based remote photoplethysmography (rPPG), we needed to extract ground truth values of electrocardiogram (ECG) and pulse oximetry (SpO2) signals from the Philips vitals monitor while recording the facial video of the subject, simultaneously. This ground truth data will be used to train the model that will perform rPPG. Various software can extract data from the Philips vitals monitor with features like data acquisition, parsing, and visualisation, but they lack synchronisation with the facial video. Therefore, we developed the Patient Monitor Data Extractor (PMDE), which collects data from the Philips IntelliVue monitors following the Data export interface programming guide provided by Philips. We set up a DHCP server on a Windows 7 computer with a webcam and interfaced with the monitor through LAN with UDP/IP. We used C++ and Windows Sockets API to develop our software and communicate over UDP. For synchronisation with the video cameras, we turned off the light in the room and used this sudden brightness drop as a trigger. The timestamp of the monitor was recorded when the webcam detected the trigger. The PMDE software records ECG at 500 Hz and SpO2 at 125 Hz with a synchronisation error of less than two sampling periods, which is about 40 ms for a 50 fps video. We conclude that PMDE is uniquely suited for recording data for rPPG evaluation because of its synchronisation feature. We have used PMDE to collect a dataset of facial videos with ground truth ECG and SpO2 signals. We intend to make PMDE available as open source to save other researchers time.

Chronic respiratory diseases are leading causes of morbidity and mortality in EU and worldwide. The current management of respiratory diseases allows only a momentary patient assessment at the time point of outpatient department visit or hospitalization. The incorporation of novel low-cost electronics in garments presents great potential for making accurate and effective continuous monitoring of lung diseases feasible. The EU WELMO project (Wearable Electronics for Effective Lung Monitoring) developed novel miniaturized sensors, integrated to a comfortable vest, enabling the accurate and continuous monitoring of the lungs, through the collection of lung sounds and EIT signals, that can be combined, processed and linked with specific clinical outcome, rendering the systematic, accurate and real-time evaluation of respiratory conditions possible. The data and features derived from lung sounds, EIT and medical sensors are best exploited in an integrated manner, towards identifying patterns of diagnostic value and patient coaching/adherence value. The proposed WELMO also offers novel algorithms for processing the collected data, and applications for presentation of the processing outcomes. Two user applications the Patient app and the HCP app were developed. This paper presents the design of WELMO Patient App. The description of software design yields the systematic process to consider every aspect and challenge of the proposed system. The presentation focuses on the functionality of Patient Application and the ease of use by health experts. The design led to a functional software which complements the hardware and is an essential and novel part of the innovative lung monitoring system of WELMO.

Portable physical activity monitors provide detailed, continuous and objective measurements of individual physical activity in the environment of daily activities. A major problem with wristbands, pedometers, and smartphones that use accelerometer technology is that they measure involuntary jerks as steps. Therefore, they generate inaccurate values ​​resulting in erroneous data. Therefore, the purpose of this study is to determine and contrast the performance obtained for the classification of daily activities of different machine learning models based on characteristics in the time domain of signals obtained from accelerometry collected at various points of the body. The development of the activity identifier is based on models and characteristics of existing developments of calorie counters by accelerometric signals; these features are extracted in the time domain. The following classifiers were applied: Logistic Regression, K-Nearest Neighbors, Support Vector Machine, Gaussian Naive Bayes, Decision Tree, Random Forest, Light Gradient Boosting and Extreme Gradient Boosting. The performance of each model was measured by how accurately it emerged to classify 4 daily activities based on the test set. The results show that, to have an accuracy greater than 70% in most models, at least 2 accelerometers are required.

Ballistocardiogram (BCG) is a measure of the ballistic force when the heart pumps blood into the circulatory system. Although the electrocardiogram (ECG) is the standard technique for diagnosis, it requires physical contact with patients’ skin. BCG systems use non-intrusive sensors (EMFi) to acquire the signal, which can be used to monitor patient vital signs in waiting rooms or other non-standard settings.ICBHI22 challenged participants to detect heartbeats from BCG records on a provided dataset acquired from patients and volunteers. Each algorithm was tested on a hidden dataset, to avoid fine-tuning. A synchronised ECG signal was used as gold-standard to score the competing algorithms. The results show the feasibility of BCG beat detection in different conditions and provide a comparison point for future developments.

Ballistocardiography (BCG) is one of various noninvasive methods used for measuring the cardiac’s activity. On the contrary to electrocardiography (ECG), ballistocardiography is a method based on the measurement of the human body motion caused by cardiac activity. BCG signals have been applied for assessing and monitoring the heart’s health. A beat detection is one of the most fundamental processing stages. In this study, a computational algorithm based on the continuous wavelet transform (CWT) is developed for the ICBHI2022 Scientific Challenge. The main goal of the ICBHI2022 Scientific Challenge is to estimate the RR intervals using a BCG signal. The performance on RR interval estimation is determined using the global score that is obtained from the difference between RR intervals estimated from detected BCG beats and actual RR intervals. The CWT-based BCG beat detection proposed is composed of two main steps: 1) CWT-based filtering; and 2) peak detection. From all ten entries in the Phase II of the ICBHI2022 Scientific Challenge, the best global score achieved using the proposed CWT-based BCG beat detection is 133.41 with the upper frequency $$f_u$$ f u , the lower frequency $$f_l$$ f l , and the minimum peak distance d of 6.6 Hz, 4.2 Hz, and 109 samples, respectively.

Ballistocardiography (BCG) signal is an alternative to classical electocardiography (ECG) in several healthcare applications, because his non-contact feature. However, signal weakness and vulnerability to multiple sources of noise due to the high sensibility acquisition is a challenge for it. In order to design and test new algorithms for BCG processing, the ICBHI 2022 held the fourth edition of the Scientific Challenge Competition for biomedical purposes such as non-invasive cardiac monitoring and signal processing of BCG. This research followed the methods based on the Pan-Tompkins algorithm for detection of QRS complexes in ECG signals with a database composed by 4 datasets of healthy volunteers (before and after excersise) and patients with atrial fibrillation (AF). Peak prominence of the signal to find local maxima was selected as the main function for processing data and comparison between BCG and ECG signals of the subjects. Score metrics established by ICBHI to assess the algorithm error showed that initial results for the first and second phase a score of 5686.2 and 3349.1 respectively, thus concluding that the algorithm proposed in the present research project based on the Scientific Challenge proved to have acceptable performance to detect heartbeats based on BCG signal datasets for both volunteers and patients subjects.

In this paper we present an algorithm for identifying individual heart beats from ballistocardiogram (BCG) records as part of IFMBE Scientific Challenge for ICBHI 2022. The main purpose of this challenge is to explore and test new algorithms of BCG signal processing for heart beat detection from BCG signal. The BCG signal records dataset included records of healthy subjects and also of patients previously diagnosed with atrial fibrillation (AF). An electrocardiogram (ECG) was acquired simultaneously for reference. The R peaks of the ECG signal were annotated as a reference, with the goal of detecting the J peaks of BCG signal which correspond to these R peaks. Our approach to this challenge consisted of two different algorithms. A comparison between using peak detection of a BCG signal only in the time domain and both in time and frequency domain using continuous wavelet transform (CWT) was made. Apart from the local maxima, the local minima were also detected, and later used for extraction of possible motion artifacts. The results showed that using CWT provided better J peak detection, and therefore better heart rate detection than regular time domain method.