Early Prediction of Neonatal Sepsis From Synthetic Clinical Data Using Machine Learning.

Neonatal sepsis is one of the most serious complications in neonatal intensive care units. Due to the often immature immune system, sepsis-related comorbidities are the major contributors to increased neonatal mortality. The rapid progression of the disease makes early treatment critical for patient survival. However, early diagnosis of sepsis remains difficult due to its non-specific symptoms. In recent years, Machine Learning-based techniques have been used in various medical applications to predict diseases using clinical data. In this work, we optimized and evaluated four prediction models with different architectural concepts. Two public datasets containing clinical data from adults and neonates were used for training. The adult data were collected to pre-train the models. Since neonatal data with sepsis diagnosis are very limited, we propose an augmentation method to generate synthetic clinical data. For the final evaluation, the real data of neonatal patients were defined as a test set. An AUROC of 0.91 and an AUPRC of 0.38 were obtained. These results are promising for early prediction of neonatal sepsis using artificial data for augmentation.Clinical relevance- This work demonstrates the potential of Machine Learning-based prediction models for the detection of sepsis to improve the early diagnosis of life-threatening conditions in neonatal intensive care units.

Spatiotemporal gait parameters in young individuals wearing an age simulation suit compared to healthy older individuals.

INTRODUCTION: Aging is accompanied by changes in muscle mass, strength and loss of sensory, visual and auditive functions. However, these changes do not occur linearly, most spatiotemporal gait parameters change with aging. Age simulation suits have been invented to give young people an impression of the implications of being older and may be a useful tool in the scientific setting for gerontology research to validate any study concept before it becomes a pilot study. The rationale behind this study was to investigate the effects of an age simulation suit on gait parameters in young healthy adults and to compare the altered gait with healthy older, community-dwelling citizens. METHODS: Subjects were 14 healthy young adults (6 female) and 8 healthy older (4 female) individuals with a mean (± SD) age of 24.8 ± 3.4 years and 72 ± 1.9 years, respectively. After initial baseline measurements had been taken and a familiarization phase, the younger subjects walked for 15 min without and 15 min with an age simulation suit on an instrumented treadmill. The older subjects walked once for 15 min on the same treadmill without wearing an age simulation suit. The walking speed was self-selected for all subjects. RESULTS: The age simulation suit reduced the walking speed from 4.1 ± 0.7 km/h to 3.3 ± 0.5 km/h (p < 0.001) in young adults with no differences compared to older adults (2.9 ± 0.6 km/h, p = 0.9). Step width increased from 8.7 ± 2.2 cm to 12.1 ± 2.2 cm (p < 0.001) and did not differ from older participants (11.1 ± 4.3 cm, p = 0.37). The stride length was reduced (132.6 ± 5.9 cm vs 118.1 +-6.6 cm, p < 0.001), but still did not match the old control group (94.5 ± 5.6 cm, p < 0.05). Wearing the suit increased thestride time of young subjects (from 1,152 to 1,316 ms, p < 0.001) and was different compared to the older control group (1,172 ms, p = 0.53). The coefficient of variation (COV) of spatiotemporal parameters did not differ between young (both not wearing the suit and wearing the suit) and older subjects. The standard deviation of lateral symmetry, an in-house marker from the instrumented treadmill that serves as a marker of gait variability, differed between young subjects without the suit and older subjects (5.89 ± 1.9 mm vs 14.6 ± 5.7 mm, p < 0.001) but not between young subjects wearing the suit and older subjects (16.4 ± 7.4 mm vs 14.6 ± 5.7 mm, p = 0.53). CONCLUSION: Wearing an age simulation suit while walking on a treadmill with a self-selected walking speed alters some, but not all, measured spatiotemporal parameters to approximate a gait pattern similar to that of an older individual.

Estimation of Step Length With Wearable Thigh Sensor Using an Unscented Kalman Filter.

The determination of step length, an important gait parameter, has been a challenging task. Although unobtrusive sensors (inertial measurement units) have been developed recently, they cannot facilitate the automatic estimation of step length. In this article, we use a model-based technique to determine the step length using the Unscented Kalman Filter with angular velocity from a gyroscope inside the thigh pocket. We then propose a novel covariance estimation algorithm based on a screening technique that performs a search for the optimal Process Noise Covariance matrix. Upon implementing the Unscented Kalman Filter, the step length is found using the horizontal position of the foot relative to the hip using a patient-independent robust peak detection algorithm. This research article paves the way for algorithms that are computationally much faster than black box methods, with more scope for the development of better algorithms for covariance estimation using the one proposed in this article as a foundation.

A Setup for Camera-Based Detection of Simulated Pathological States Using a Neonatal Phantom.

Premature infants are among the most vulnerable patients in a hospital. Due to numerous complications associated with immaturity, a continuous monitoring of vital signs with a high sensitivity and accuracy is required. Today, wired sensors are attached to the patient's skin. However, adhesive electrodes can be potentially harmful as they can damage the very thin immature skin. Although unobtrusive monitoring systems using cameras show the potential to replace cable-based techniques, advanced image processing algorithms are data-driven and, therefore, need much data to be trained. Due to the low availability of public neonatal image data, a patient phantom could help to implement algorithms for the robust extraction of vital signs from video recordings. In this work, a camera-based system is presented and validated using a neonatal phantom, which enabled a simulation of common neonatal pathologies such as hypo-/hyperthermia and brady-/tachycardia. The implemented algorithm was able to continuously measure and analyze the heart rate via photoplethysmography imaging with a mean absolute error of 0.91 bpm, as well as the distribution of a neonate's skin temperature with a mean absolute error of less than 0.55 °C. For accurate measurements, a temperature gain offset correction on the registered image from two infrared thermography cameras was performed. A deep learning-based keypoint detector was applied for temperature mapping and guidance for the feature extraction. The presented setup successfully detected several levels of hypo- and hyperthermia, an increased central-peripheral temperature difference, tachycardia and bradycardia.

Model-Based Estimation of Ankle Joint Stiffness.

We address the estimation of biomechanical parameters with wearable measurement technologies. In particular, we focus on the estimation of sagittal plane ankle joint stiffness in dorsiflexion/plantar flexion. For this estimation, a novel nonlinear biomechanical model of the lower leg was formulated that is driven by electromyographic signals. The model incorporates a two-dimensional kinematic description in the sagittal plane for the calculation of muscle lever arms and torques. To reduce estimation errors due to model uncertainties, a filtering algorithm is necessary that employs segmental orientation sensor measurements. Because of the model's inherent nonlinearities and nonsmooth dynamics, a square-root cubature Kalman filter was developed. The performance of the novel estimation approach was evaluated in silico and in an experimental procedure. The experimental study was conducted with body-worn sensors and a test-bench that was specifically designed to obtain reference angle and torque measurements for a single joint. Results show that the filter is able to reconstruct joint angle positions, velocities and torque, as well as, joint stiffness during experimental test bench movements.

Observer-Based Human Knee Stiffness Estimation.

OBJECTIVE: We consider the problem of stiffness estimation for the human knee joint during motion in the sagittal plane. METHODS: The new stiffness estimator uses a nonlinear reduced-order biomechanical model and a body sensor network (BSN). The developed model is based on a two-dimensional knee kinematics approach to calculate the angle-dependent lever arms and the torques of the muscle-tendon-complex. To minimize errors in the knee stiffness estimation procedure that result from model uncertainties, a nonlinear observer is developed. The observer uses the electromyogram (EMG) of involved muscles as input signals and the segmental orientation as the output signal to correct the observer-internal states. Because of dominating model nonlinearities and nonsmoothness of the corresponding nonlinear functions, an unscented Kalman filter is designed to compute and update the observer feedback (Kalman) gain matrix. RESULTS: The observer-based stiffness estimation algorithm is subsequently evaluated in simulations and in a test bench, specifically designed to provide robotic movement support for the human knee joint. CONCLUSION: In silico and experimental validation underline the good performance of the knee stiffness estimation even in the cases of a knee stiffening due to antagonistic coactivation. SIGNIFICANCE: We have shown the principle function of an observer-based approach to knee stiffness estimation that employs EMG signals and segmental orientation provided by our own IPANEMA BSN. The presented approach makes realtime, model-based estimation of knee stiffness with minimal instrumentation possible.

Linearity of electrical impedance tomography during maximum effort breathing and forced expiration maneuvers.

Electrical impedance tomography (EIT) provides global and regional information about ventilation by means of relative changes in electrical impedance measured with electrodes placed around the thorax. In combination with lung function tests, e.g. spirometry and body plethysmography, regional information about lung ventilation can be achieved. Impedance changes strictly correlate with lung volume during tidal breathing and mechanical ventilation. Initial studies presumed a correlation also during forced expiration maneuvers. To quantify the validity of this correlation in extreme lung volume changes during forced breathing, a measurement system was set up and applied on seven lung-healthy volunteers. Simultaneous measurements of changes in lung volume using EIT imaging and pneumotachography were obtained with different breathing patterns. Data was divided into a synchronizing phase (spontaneous breathing) and a test phase (maximum effort breathing and forced maneuvers). The EIT impedance changes correlate strictly with spirometric data during slow breathing with increasing and maximum effort ([Formula: see text]) and during forced expiration maneuvers ([Formula: see text]). Strong correlations in spirometric volume parameters [Formula: see text] ([Formula: see text]), [Formula: see text]/FVC ([Formula: see text]), and flow parameters PEF, [Formula: see text], [Formula: see text], [Formula: see text] ([Formula: see text]) were observed. According to the linearity during forced expiration maneuvers, EIT can be used during pulmonary function testing in combination with spirometry for visualisation of regional lung ventilation.

Body-Sensor-Network-Based Spasticity Detection.

Spasticity is a common disorder of the skeletal muscle with a high incidence in industrialised countries. A quantitative measure of spasticity using body-worn sensors is important in order to assess rehabilitative motor training and to adjust the rehabilitative therapy accordingly. We present a new approach to spasticity detection using the Integrated Posture and Activity Network by Medit Aachen body sensor network (BSN). For this, a new electromyography (EMG) sensor node was developed and employed in human locomotion. Following an analysis of the clinical gait data of patients with unilateral cerebral palsy, a novel algorithm was developed based on the idea to detect coactivation of antagonistic muscle groups as observed in the exaggerated stretch reflex with associated joint rigidity. The algorithm applies a cross-correlation function to the EMG signals of two antagonistically working muscles and subsequent weighting using a Blackman window. The result is a coactivation index which is also weighted by the signal equivalent energy to exclude positive detection of inactive muscles. Our experimental study indicates good performance in the detection of coactive muscles associated with spasticity from clinical data as well as measurements from a BSN in qualitative comparison with the Modified Ashworth Scale as classified by clinical experts. Possible applications of the new algorithm include (but are not limited to) use in robotic sensorimotor therapy to reduce the effect of spasticity.

Multi-sensor calibration of low-cost magnetic, angular rate and gravity systems.

We present a new calibration procedure for low-cost nine degrees-of-freedom (9DOF) magnetic, angular rate and gravity (MARG) sensor systems, which relies on a calibration cube, a reference table and a body sensor network (BSN). The 9DOF MARG sensor is part of our recently-developed "Integrated Posture and Activity Network by Medit Aachen" (IPANEMA) BSN. The advantage of this new approach is the use of the calibration cube, which allows for easy integration of two sensor nodes of the IPANEMA BSN. One 9DOF MARG sensor node is thereby used for calibration; the second 9DOF MARG sensor node is used for reference measurements. A novel algorithm uses these measurements to further improve the performance of the calibration procedure by processing arbitrarily-executed motions. In addition, the calibration routine can be used in an alignment procedure to minimize errors in the orientation between the 9DOF MARG sensor system and a motion capture inertial reference system. A two-stage experimental study is conducted to underline the performance of our calibration procedure. In both stages of the proposed calibration procedure, the BSN data, as well as reference tracking data are recorded. In the first stage, the mean values of all sensor outputs are determined as the absolute measurement offset to minimize integration errors in the derived movement model of the corresponding body segment. The second stage deals with the dynamic characteristics of the measurement system where the dynamic deviation of the sensor output compared to a reference system is Sensors 2015, 15 25920 corrected. In practical validation experiments, this procedure showed promising results with a maximum RMS error of 3.89°.