Early detection of late-onset neonatal sepsis from noninvasive biosignals using deep learning: A multicenter prospective development and validation study.

BACKGROUND: Neonatal sepsis is responsible for significant morbidity and mortality worldwide. Its accurate and timely diagnosis is hindered by vague symptoms and the urgent necessity for early antibiotic intervention. The gold standard for diagnosing the condition is the identification of a pathogenic organism from normally sterile sites via laboratory testing. However, this method is resource-intensive and cannot be conducted continuously. OBJECTIVE: This study aimed to predict the onset of late-onset sepsis (LOS) with good diagnostic value as early as possible using non-invasive biosignal measurements from neonatal intensive care unit (NICU) monitors. METHODS: In this prospective multicenter study, we developed a multimodal machine learning algorithm based on a convolutional neural network (CNN) structure that uses the power spectral density (PSD) of recorded biosignals to predict the onset of LOS. This approach aimed to discern LOS-related pathogenic spectral signatures without labor-intensive manual artifact removal. RESULTS: The model achieved an area under the receiver operating characteristic score of 0.810 (95 % CI 0.698-0.922) on the validation dataset. With an optimal operating point, LOS detection had 83 % sensitivity and 73 % specificity. The median early detection was 44 h before clinical suspicion. The results highlighted the additive importance of electrocardiogram and respiratory impedance (RESP) signals in improving predictive accuracy. According to a more detailed analysis, the predictive power arose from the morphology of the electrocardiogram's R-wave and sudden changes in the RESP signal. CONCLUSION: Raw biosignals from NICU monitors, in conjunction with PSD transformation, as input to the CNN, can provide state-of-the-art prediction performance for LOS without the need for artifact removal. To the knowledge of the authors, this is the first study to highlight the independent and additive predictive potential of electrocardiogram R-wave morphology and concurrent, sudden changes in the RESP waveform in predicting the onset of LOS using non-invasive biosignals.

What Makes a Quality Health App-Developing a Global Research-Based Health App Quality Assessment Framework for CEN-ISO/TS 82304-2: Delphi Study.

BACKGROUND: The lack of an international standard for assessing and communicating health app quality and the lack of consensus about what makes a high-quality health app negatively affect the uptake of such apps. At the request of the European Commission, the international Standard Development Organizations (SDOs), European Committee for Standardization, International Organization for Standardization, and International Electrotechnical Commission have joined forces to develop a technical specification (TS) for assessing the quality and reliability of health and wellness apps. OBJECTIVE: This study aimed to create a useful, globally applicable, trustworthy, and usable framework to assess health app quality. METHODS: A 2-round Delphi technique with 83 experts from 6 continents (predominantly Europe) participating in one (n=42, 51%) or both (n=41, 49%) rounds was used to achieve consensus on a framework for assessing health app quality. Aims included identifying the maximum 100 requirement questions for the uptake of apps that do or do not qualify as medical devices. The draft assessment framework was built on 26 existing frameworks, the principles of stringent legislation, and input from 20 core experts. A follow-up survey with 28 respondents informed a scoring mechanism for the questions. After subsequent alignment with related standards, the quality assessment framework was tested and fine-tuned with manufacturers of 11 COVID-19 symptom apps. National mirror committees from the 52 countries that participated in the SDO technical committees were invited to comment on 4 working drafts and subsequently vote on the TS. RESULTS: The final quality assessment framework includes 81 questions, 67 (83%) of which impact the scores of 4 overarching quality aspects. After testing with people with low health literacy, these aspects were phrased as "Healthy and safe," "Easy to use," "Secure data," and "Robust build." The scoring mechanism enables communication of the quality assessment results in a health app quality score and label, alongside a detailed report. Unstructured interviews with stakeholders revealed that evidence and third-party assessment are needed for health app uptake. The manufacturers considered the time needed to complete the assessment and gather evidence (2-4 days) acceptable. Publication of CEN-ISO/TS 82304-2:2021 Health software - Part 2: Health and wellness apps - Quality and reliability was approved in May 2021 in a nearly unanimous vote by 34 national SDOs, including 6 of the 10 most populous countries worldwide. CONCLUSIONS: A useful and usable international standard for health app quality assessment was developed. Its quality, approval rate, and early use provide proof of its potential to become the trusted, commonly used global framework. The framework will help manufacturers enhance and efficiently demonstrate the quality of health apps, consumers, and health care professionals to make informed decisions on health apps. It will also help insurers to make reimbursement decisions on health apps.

Development Towards Patient-Centered eHealth Services in Finland.

The paper analyses the development of public eHealth services from 2014 to 2021 from the patients' point of view. The merits and missing features of the eHealth services were identified with patient interviews in 2014-2015. The list of missing features was again checked against the eHealth services in 2021. The main finding was that all the features wanted by the patients had still not been implemented. The finding of this paper suggests that current Finnish public eHealth services are organizations oriented rather than patient oriented.

The Definition of Informatics Competencies in Finnish Healthcare and Social Welfare Education.

Finland is a world leader in the use of public electronic services. Continuous improvement to competencies is a prerequisite for the success of digitalisation in the service development sector. The increasing use of information technology in health and social care needs to be taken into account in the education of the health and social care sector work force. The mandate of the national SotePeda 24/7 project is to identify and define the informatics competencies required for multidisciplinary education of this sector in Finland. The project has adapted international recommendations for use in the national context. The national recommendation covers 12 areas of competency and related content. In addition to defining competencies, the project has produced a toolbox of materials for use by educators of these topics in universities that cover applied sciences and lifelong learning. The results of the project are expected to significantly improve the preparedness of graduating health and social care and related engineering and business sector students to make full use information technology, all of which benefits the national health and social welfare system.

IEC 62304 Ed. 2: Software Life Cycle Standard for Health Software.

The quality of software is high in medical devices due to the strict regulatory requirements and their implementation in the software development processes through the use of the IEC 62304 standard. The goal of this standard revision project was to extend the scope of the standard to all health software and also to bring the requirements of the 12 year old standard back to the state-of-the-art including provisions for cybersecurity. The joint IEC/SC62A and ISO/TC215 project team revised the standard and adapted its risk management, usability, and security requirements to serve both the medical device industry and the overall health software industry. The resulting second version of the standard has gone through a multistage global voting process to achieve a consensus of the requirements to serve both these communities. The resulting standard has potential to have a major impact on the quality of software used in health care globally.

Detection of Snores Using Source Separation on an Emfit Signal.

Snoring (SN) is an early sign of upper airway dysfunction, and it is strongly associated with obstructive sleep apnea. SN detection is important to monitor SN objectively and to improve the diagnostic sensitivity of sleep-disordered breathing. In this study, an automatic snore detection method using an electromechanical film transducer (Emfit) signal is presented. Representative polysomnographs of normal breathing and SN periods from 30 subjects were selected. Individual SN events were identified using source separation applying nonnegative matrix factorization deconvolution. The algorithm was evaluated using manual annotation of the polysomnographic recordings. According to our results, the sensitivity, and the positive predictive value of the developed method to reveal snoring from the Emfit signal were 82.81% and 86.29%, respectively. Compared to other approaches, our method adapts to the individual spectral snoring profile of the subject rather than matching a particular spectral profile, estimates the snoring intensity, and obtains the specific spectral profile of the snores in the epoch. Additionally, no training is necessary. This study suggests that it is possible to detect individual SN events with Emfit mattress, which can be used as a contactless alternative to more conventional methods such as piezo-snore sensors or microphones.

Assessment of support vector machines and convolutional neural networks to detect snoring using Emfit mattress.

Snoring (SN) is an essential feature of sleep breathing disorders, such as obstructive sleep apnea (OSA). In this study, we evaluate epoch-based snoring detection methods using an unobtrusive electromechanical film transducer (Emfit) mattress sensor using polysomnography recordings as a reference. Two different approaches were investigated: a support vector machine (SVM) classifier fed with a subset of spectral features and convolutional neural network (CNN) fed with spectrograms. Representative 10-min normal breathing (NB) and SN periods were selected for analysis in 30 subjects and divided into thirty-second epochs. In the evaluation, average results over 10 fold Monte Carlo cross-validation with 80% training and 20% test split were reported. Highest performance was achieved using CNN, with 92% sensitivity, 96% specificity, 94% accuracy, and 0.983 area under the receiver operating characteristics curve (AROC). Results showed a 6% average increase of performance of the CNN over SVM and greater robustness, and similar performance to ambient microphones.

Spectral analysis of snoring events from an Emfit mattress.

The aim of this study is to explore the capability of an Emfit (electromechanical film transducer) mattress to detect snoring (SN) by analyzing the spectral differences between normal breathing (NB) and SN. Episodes of representative NB and SN of a maximum of 10 min were visually selected for analysis from 33 subjects. To define the bands of interest, we studied the statistical differences in the power spectral density (PSD) between both breathing types. Three bands were selected for further analysis: 6-16 Hz (BW1), 16-30 Hz (BW2) and 60-100 Hz (BW3). We characterized the differences between NB and SN periods in these bands using a set of spectral features estimated from the PSD. We found that 15 out of the 29 features reached statistical significance with the Mann-Whitney U-test. Diagnostic properties for each feature were assessed using receiver operating characteristic analysis. According to our results, the highest diagnostic performance was achieved using the power ratio between BW2 and BW3 (0.85 area under the receiver operating curve, 80% sensitivity, 80% specificity and 80% accuracy). We found that there are significant differences in the defined bands between the NB and SN periods. A peak was found in BW3 for SN epochs, which was best detected using power ratios. Our work suggests that it is possible to detect snoring with an Emfit mattress. The mattress-type movement sensors are inexpensive and unobtrusive, and thus provide an interesting tool for sleep research.

Sleep stage classification with low complexity and low bit rate.

Standard sleep stage classification is based on visual analysis of central (usually also frontal and occipital) EEG, two-channel EOG, and submental EMG signals. The process is complex, using multiple electrodes, and is usually based on relatively high (200-500 Hz) sampling rates. Also at least 12 bit analog to digital conversion is recommended (with 16 bit storage) resulting in total bit rate of at least 12.8 kbit/s. This is not a problem for in-house laboratory sleep studies, but in the case of online wireless self-applicable ambulatory sleep studies, lower complexity and lower bit rates are preferred. In this study we further developed earlier single channel facial EMG/EOG/EEG-based automatic sleep stage classification. An algorithm with a simple decision tree separated 30 s epochs into wakefulness, SREM, S1/S2 and SWS using 18-45 Hz beta power and 0.5-6 Hz amplitude. Improvements included low complexity recursive digital filtering. We also evaluated the effects of a reduced sampling rate, reduced number of quantization steps and reduced dynamic range on the sleep data of 132 training and 131 testing subjects. With the studied algorithm, it was possible to reduce the sampling rate to 50 Hz (having a low pass filter at 90 Hz), and the dynamic range to 244 microV, with an 8 bit resolution resulting in a bit rate of 0.4 kbit/s. Facial electrodes and a low bit rate enables the use of smaller devices for sleep stage classification in home environments.

Using the nonlinear control of anaesthesia-induced hypersensitivity of EEG at burst suppression level to test the effects of radiofrequency radiation on brain function.

BACKGROUND: In this study, investigating the effects of mobile phone radiation on test animals, eleven pigs were anaesthetised to the level where burst-suppression pattern appears in the electroencephalogram (EEG). At this level of anaesthesia both human subjects and animals show high sensitivity to external stimuli which produce EEG bursts during suppression. The burst-suppression phenomenon represents a nonlinear control system, where low-amplitude EEG abruptly switches to very high amplitude bursts. This switching can be triggered by very minor stimuli and the phenomenon has been described as hypersensitivity. To test if also radio frequency (RF) stimulation can trigger this nonlinear control, the animals were exposed to pulse modulated signal of a GSM mobile phone at 890 MHz. In the first phase of the experiment electromagnetic field (EMF) stimulation was randomly switched on and off and the relation between EEG bursts and EMF stimulation onsets and endpoints were studied. In the second phase a continuous RF stimulation at 31 W/kg was applied for 10 minutes. The ECG, the EEG, and the subcutaneous temperature were recorded. RESULTS: No correlation between the exposure and the EEG burst occurrences was observed in phase I measurements. No significant changes were observed in the EEG activity of the pigs during phase II measurements although several EEG signal analysis methods were applied. The temperature measured subcutaneously from the pigs' head increased by 1.6 degrees C and the heart rate by 14.2 bpm on the average during the 10 min exposure periods. CONCLUSION: The hypothesis that RF radiation would produce sensory stimulation of somatosensory, auditory or visual system or directly affect the brain so as to produce EEG bursts during suppression was not confirmed.

Automatic sleep detection using activity and facial electrodes.

The most commonly applied unobtrusive sleep monitoring method is actigraphy, the measurement of body limb movements. In spite of its wide clinical acceptance, actigraphy has a low specificity for sleep detection leaving room for novel approaches of unobtrusive sleep monitoring. The present study compared sleep detection by a novel single channel electro-oculography (EOG) method and three activity monitors, with the golden standard of polysomnographic sleep analysis as a reference. With standard actigraphy (Actiwatch placed at the left wrist) sleep detection specificity and sensitivity were 42% and 95%. With the Alive Monitor attached on the same wrist, activity-based sleep detection specificity and sensitivity were 40% and 97%. With another Alive Monitor placed over the sternum sleep detection specificity and sensitivity were 21% and 99%. With two self-applied EOG electrodes combined with automatic sleep detection analysis, specificity and sensitivity were 72% and 96%. The results confirm low specificity of actigraphic sleep estimates, and demonstrate that the novel single-channel EOG method provides a substantial improvement in specificity.

Automatic sleep stage classification using two facial electrodes.

Standard sleep stage classification is based on visual analysis of central EEG, EOG and EMG signals. Automatic analysis with a reduced number of sensors has been studied as an easy alternative to the standard. In this study, a single-channel electro-oculography (EOG) algorithm was developed for separation of wakefulness, SREM, light sleep (S1, S2) and slow wave sleep (S3, S4). The algorithm was developed and tested with 296 subjects. Additional validation was performed on 16 subjects using a low weight single-channel Alive Monitor. In the validation study, subjects attached the disposable EOG electrodes themselves at home. In separating the four stages total agreement (and Cohen's Kappa) in the training data set was 74% (0.59), in the testing data set 73% (0.59) and in the validation data set 74% (0.59). Self-applicable electro-oculography with only two facial electrodes was found to provide reasonable sleep stage information.

Reducing the effects of electrocardiographic artifacts on electro-oculography in automatic sleep analysis.

Removal of electrocardiographic (ECG) artifacts of QRS complexes from a single channel electroencephalography (EEG) and electro-oculography (EOG) can be problematic especially when no reference ECG signal is available. This study examined a simple estimation method excluding the possible QRS part of the EOG trace before spectrum estimation. The method was tested using a simple sleep classifier based on 0.5-30 Hz mean frequency of single channel sleep EOG, with the left EOG electrode referenced to the left mastoid (EOG L-M1). When QRS peaks were automatically excluded from the least square (LS) mean frequency estimation the average optimal mean frequency threshold decreased from 9.3 Hz to 8.8 Hz and agreement and Cohen's Kappa increased respectively from 89% to 90% and from 0.44 to 0.50 when compared to the traditional spectral estimation.

Automatic sleep stage classification using two-channel electro-oculography.

An automatic method for the classification of wakefulness and sleep stages SREM, S1, S2 and SWS was developed based on our two previous studies. The method is based on a two-channel electro-oculography (EOG) referenced to the left mastoid (M1). Synchronous electroencephalographic (EEG) activity in S2 and SWS was detected by calculating cross-correlation and peak-to-peak amplitude difference in the 0.5-6 Hz band between the two EOG channels. An automatic slow eye-movement (SEM) estimation was used to indicate wakefulness, SREM and S1. Beta power 18-30 Hz and alpha power 8-12 Hz was also used for wakefulness detection. Synchronous 1.5-6 Hz EEG activity and absence of large eye movements was used for S1 separation from SREM. Simple smoothing rules were also applied. Sleep EEG, EOG and EMG were recorded from 265 subjects. The system was tuned using data from 132 training subjects and then applied to data from 131 validation subjects that were different to the training subjects. Cohen's Kappa between the visual and the developed new automatic scoring in separating 30s wakefulness, SREM, S1, S2 and SWS epochs was substantial 0.62 with epoch by epoch agreement of 72%. With automatic subject specific alpha thresholds for offline applications results improved to 0.63 and 73%. The automatic method can be further developed and applied for ambulatory sleep recordings by using only four disposable, self-adhesive and self-applicable electrodes.

Development and comparison of four sleep spindle detection methods.

OBJECTIVE: The objective of the present work was to develop and compare methods for automatic detection of bilateral sleep spindles. METHODS AND MATERIALS: All-night sleep electroencephalographic (EEG) recordings of 12 healthy subjects with a median age of 40 years were studied. The data contained 6043 visually scored bilateral spindles occurring in frontopolar or central brain location. In the present work a new sigma index for spindle detection was developed, based on the fast Fourier transform (FFT) spectrum, aiming at approximating our previous fuzzy spindle detector. The sigma index was complemented with spindle amplitude analysis, based on finite impulse response (FIR) filtering, to form of a combination detector of bilateral spindles. In this combination detector, the spindle amplitude distribution of each recording was estimated and used to tune two different amplitude thresholds. This combination detector was compared to bilaterally extracted sigma indexes and fuzzy detections, which aim to be independent of absolute spindle amplitudes. As a fourth method a fixed spindle amplitude detector was included. RESULTS: The combination detector provided the best overall performance; in S2 sleep a 70% true positive rate was reached with a specificity of 98.6%, and a false-positive rate of 32%. The bilateral sigma indexes provided the second best results, followed by fuzzy detector, while the fixed amplitude detector provided the poorest results so that in S2 sleep a 70% true positive rate was reached with a specificity of 97.7% and false-positive rate of 46%. The spindle amplitude distributions automatically determined for each recording by the combination detector were compared to amplitudes of visually scored spindles and they proved to correspond well. Inter-hemispheric amplitude variation of visually scored bilateral spindles is also presented. CONCLUSION: Flexibility is beneficial in the detection of bilateral spindles. The present work advances automated spindle detection and increases the knowledge of bilateral sleep spindle characteristics.

The use of two-channel electro-oculography in automatic detection of unintentional sleep onset.

An automatic method was developed for detecting unintentional sleep onset. The automatic method is based on a two-channel electro-oculography (EOG) with left mastoid (M1) as reference. An automatic estimation of slow eye movements (SEM) was developed and used as the main criterion to separate sleep stage 1 (S1) from wakefulness. Additionally synchronous electroencephalographic (EEG) activity of sleep stages 1 and 2 was detected by calculating cross-correlation and amplitude difference in the 1.5-6 Hz theta band between the two EOG channels. Alpha power 8-12 Hz and beta power 18-30 Hz were used to determine wakefulness. Unintentional sleep onsets were studied using data from four separate maintenance of wakefulness test (MWT) sessions of 228 subjects. The automatic scoring of 30s sleep onset epochs using only EOG was compared to standard visual sleep stage scoring. The optimal detection thresholds were derived using data from 114 subjects and then applied to the data from different 114 subjects. Cohen's Kappa between the visual and the new automatic scoring system in separating wakefulness and sleep was substantial (0.67) with epoch by epoch agreement of 98%. The sleep epoch detection sensitivity was 70% and specificity 99%. The results are provided with a 1s delay for each 30s epoch. The developed method has to be tested in field applications. The advantage of the automatic method is that it could be applied during online recordings using only four disposable self-adhesive self-applicable electrodes.

An intelligent ballistocardiographic chair using a novel SF-ART neural network and Biorthogonal wavelets.

This paper presents a comparative analysis of novel supervised fuzzy adaptive resonance theory (SF-ART), multilayer perceptron (MLP) and Multi Layer Perceptrons (MLP) neural networks over Ballistocardiogram (BCG) signal recognition. To extract essential features of the BCG signal, we applied Biorthogonal wavelets. SF-ART performs classification on two levels. At first level, pre-classifier which is self-organized fuzzy ART tuned for fast learning classifies the input data roughly to arbitrary (M) classes. At the second level, post-classification level, a special array called Affine Look-up Table (ALT) with M elements stores the labels of corresponding input samples in the address equal to the index of fuzzy ART winner. However, in running (testing) mode, the content of an ALT cell with address equal to the index of fuzzy ART winner output will be read. The read value declares the final class that input data belongs to. In this paper, we used two well-known patterns (IRIS and Vowel data) and a medical application (Ballistocardiogram data) to evaluate and check SF-ART stability, reliability, learning speed and computational load. Initial tests with BCG from six subjects (both healthy and unhealthy people) indicate that the SF-ART is capable to perform with a high classification performance, high learning speed (elapsed time for learning around half second), and very low computational load compared to the well-known neural networks such as MLP which needs minutes to learn the training material. Moreover, to extract essential features of the BCG signal, we applied Biorthogonal wavelets. The applied wavelet transform requires no prior knowledge of the statistical distribution of data samples.

Automatic detection of slow wave sleep using two channel electro-oculography.

An automatic method was developed for detecting slow wave sleep (SWS). The automatic method is based on a two-channel electro-oculography (EOG) with left mastoid (M1) as reference. Synchronous electroencephalographic (EEG) activity was detected by calculating cross-correlation between the two EOG channels by using 0.5-6 Hz band. An amplitude criterion was used for detecting slow waves and beta power 18-30 Hz was used to exclude artefacts. The automatic scoring was compared to a standard visual sleep scoring based on EOG, central EEG and submental EMG. Sleep EEG and EOG were recorded from 265 subjects. The optimal cross-correlation, amplitude and beta thresholds were derived using data from 133 training subjects and then applied to the data from different 132 validation subjects. Results were most sensitive to the changes in the amplitude criteria. Cohen's Kappa between the visual and the new developed automatic scoring in separating non-SWS and SWS was substantial (0.70) with epoch-by-epoch agreement of 93%. SWS epoch detection sensitivity was 75% and specificity was 96%. Also the total amount of slow waves, slow wave time (SWT), was estimated. The advantage of the automatic method is that it could be applied during online recordings using only four disposable self-adhesive electrodes.

Systematic performance evaluation of a continuous-scale sleep depth measure.

In this article, systematic performance evaluation of a continuous-scale sleep depth measure will be discussed. Our main objective has been to select the adjustable analysis parameters such that the best possible correspondence between method output and standard visual sleep staging could be achieved. Sleep depth estimation was based on continuous monitoring of short-time EEG synchronization through the local mean frequency of the EEG. During the experiments, total amount of 752 different combinations of four adjustable parameters were compared based on all-night sleep EEG recordings of 15 healthy subjects. Optimization strategy applied was based on maximizing the weighted average of pair-wise separabilities of EEG mean frequency distributions in all the standard sleep stage pairs. Finally, robustness of the optimized parameters was verified with an independent dataset of 34 all-night sleep recordings. Our results show that clear topological differences between brain hemispheres and different electrode locations exist. Performance improvements of even 20-30% units can be achieved by proper selection of analysis parameters and the EEG derivation used for the analysis. Remarkable independence of system performance on the analysis window length leads to improved temporal resolution compared to that achieved through standard visual analysis. In addition to giving practical suggestions on the parameter selection, we also propose a possible method for improving stage separability especially between S2 and REM.

Apnea patients show a frontopolar inter-hemispheric spindle frequency difference.

Sleep apnea syndrome is known to disturb sleep. The purpose of the present work was to study spindle frequency in apnea patients. All-night sleep EEG recordings of 15 apnea patients and 15 control subjects with median ages of 47 and 46 years, respectively, were studied. A previously presented and validated multi-channel spindle analysis method was applied for automatic detection and frequency analysis of bilateral frontopolar and central spindles. Bilateral frontopolar spindles of apnea patients were found to show lower frequencies on the left hemisphere than on the right. Such an inter-hemispheric spindle frequency difference in apnea patients is a novel finding. It could be that the hypoxias and hypercapnias caused by apneic episodes result in local disruption in the regulation of sleep in the frontal lobes.