Abstract Abnormal behaviour can be an indicator for a medical condition in older adults. Our novel unsupervised statistical concept drift detection approach uses variational autoencoders for estimating the parameters for a statistical hypothesis test for abnormal days. As feature, the Kullback–Leibler divergence of activity probability maps derived from power and motion sensors were used. We showed the general feasibility (min. F 1 -Score of 91 %) on an artificial dataset of four concept drift types. Then we applied our new method to our real–world dataset collected from the homes of 20 (pre–)frail older adults (avg. age 84.75 y). Our method was able to find abnormal days when a participant suffered from severe medical condition.

Many physically demanding tasks require the help of a second person to avoid the otherwise harmful effects of musculoskeletal disorders when performing the task alone. The workforce in many domains, such as nursing care, is rapidly declining. Furthermore, complex repositioning tasks during patient care make it all the more difficult to find a simple solution. Therefore, robotic assistance could be a perfect fit to support this scenario. For this purpose, we conducted a study where 12 caregivers were given the task of repositioning a patient simulator (80 kg) within a nursing bed with the help of a robotic assistant. We measured the physical relief potential using muscle activity and force data while examining this human-robot collaboration problem with two different control approaches, namely a virtual reality-based and a rule-based robot control method. The results show that both methods effectively supported the chosen task with varied strengths and weaknesses. The main advantage of VR-based control is the ability to adapt to the situation and execute quickly (6.61 s), but this also results in higher acting robot forces (35.70 N). With the rule-based control variant, a lower speed of support can be noted on average (10.84 s), but better robot interaction forces (14.95 N) can be discovered in return.

Whole-arm manipulation (WAM) is often used to manipulate large and bulky objects. Contact point-based methods for generating the robot configurations for WAM mostly search for suitable contact points and configurations simultaneously. However, in order to learn good contact points, or allow an operator to select them, inverse kinematics (IK) solvers are needed which take such points along with a surface normal into account. Therefore, we propose a geometry-based IK method for WAM generate robot configurations in contact with two specific points. The method is compared against two other methods, one numerical method and a commercial IK solver. We show our method to be the fastest.

Although digitization and artificial intelligence are already being used in many areas, nutritional counseling for frailty patients is largely retrospective and based on analog questionnaires. In order to enable a broad spectrum of frailty patients to independently keep a digital nutrition diary, this paper presents a hybrid input method based on object detection using artificial intelligence in combination with a dynamic and interactive interview mode. The interview mode dynamically queries for possible missing inputs based on the objects detected by the artificial intelligence. The artificial intelligence was trained with open source and specifically for this use case generated data. A study with 21 subjects compared the hybrid approach with four other approaches and mobile applications based on usability, time effort, and detection accuracy. Especially in the area of usability, the hybrid approach came out on top.

Robotic assistance in nursing is becoming increasingly important due to labor shortage and novel capabilities in hard- and software. In nursing, patients often need to be manipulated which requires force distribution beyond the scope of traditional end effectors. This can be solved with whole arm manipulation (WAM). Unfortunately, commanding WAM configurations for grasping patients is very complex, as many contact points need to be achieved, and all 6 or 7 degrees of freedom have to be defined. In this paper, a novel interaction concept and according to inverse kinematic solvers based on a local elliptic approximation are proposed, which allows commanding a WAM holding grasp with one click on a point cloud. It is shown how the local elliptic model is used for formulating a very fast analytic multi-contact inverse kinematic solver. Two alternative solvers are derived by extending the BioIK method and by utilizing the popular sequential quadratic programming approach. Experiments show the efficiency of the analytic method and the robustness of the local elliptic model relative to perceptual perturbations.

A person’s stool can tell a lot about his or her state of health. Among other things, diarrhea or constipation lead to a reduced digestive efficiency. For many people their own stool is a shameful topic. However, the effectiveness of digestion of food has a direct influence on the recommendations for patients undergoing a nutritional therapy. This paper outlines a prototypical system for an automatic and ambient classification of stool forms into three classes: thin, normal, and hard stool based on the Bristol Stool Scale. The stool is recorded in transit after exiting the anus until it reaches the toilet floor to avoid the problems of conventional procedures. Corresponding data were generated under laboratory conditions. Various algorithms from the field of machine learning and deep learning were applied to this data. The evaluation results show that two out of five of these algorithms achieve classification rates of 100%.

As there is no standardized food database with all products available in Europe, many developers of health apps fall back on databases of communities whose quality is often insufficient. In health apps, the quality of the data sets is critical, as poor quality lowers the user's confidence. This paper examines the plausibility of nutrient data from such data sources using similarity analysis, decision support methods and Big Data technology. During a special developed process, the plausibility of the data is to be increased. Finally, the methods used will be evaluated on the basis of test data.

To simplify the evaluation and certification process for mobile medical applications, a methodology for simple testing has been developed. For this purpose, certain actions of the apps are provided with timestamps in order to be able to reproduce the entire behavior of the testers within the scope of the test study. Therefore, the tester does not have to come to a test laboratory to fulfill certain tasks that are set by the test leader. Nevertheless, all main and secondary functions are tested naturally, as is the entire app.With the help of these asynchronous remote tests, the testers can test the medical app in their familiar environment. Thus the test environment is not present and the testers act naturally. The timestamps are not noticed by the testers and require only little additional programming effort. The results are automatically transferred to a database of the developer and evaluated in terms of usability, compliance, risk factors, learning curve and validity. It is possible to automatically compare the timestamp with a previously defined Happy Path and other previously defined reference values.

Caregivers and relatives of artificially ventilated persons are under high pressure as errors or miscues in care activities can lead to the death of the patients. Therefore, the joint research project "MeSiB" aims at implementing a safety and alerting system, which analyzes the sensor events, infers whether the situation is dangerous, and, if so, alarms the caregiver and/or a telemedicine station. Among other factors, the number of persons in the bed area of the patient is an essential information for this system. Therefore, a subsystem of this safety system detects the number of persons with an infrared array sensor (Panasonic Grid-EYE). This subsystem is presented in this work. The paper describes the general approach and system design, and how the subsystem integrates into the MeSiB architecture. The structure of the software is presented, and different machine learning approaches are tested. Finally, the results of an evaluation study in a realistic home care environment are presented and discussed. In this evaluation study, the Random Forest classifier was used and detection rates of 84% and 96% for two different scenarios were achieved.

This paper focuses on integrating food data sources into a central database using extract, transform and load processing and the subsequent data quality enhancement. The obtained data will be transmitted by a food data web service to certain health apps for further use. Furthermore, it is planned to identify inconsistent, incorrect, duplicate and incomplete data using methods of data profiling so that they can be corrected. In order to quantify the data quality purposefully and appropriately, certain quality metrics were used. These metrics were calculated and evaluated using random test data selected from the food data.

The goal of the DiDiER project is to verifiably improve services in the field of dietary counselling. This will be achieved by digitising information to increase counselling intensity and to improve workflows for the service provider. The project will develop an IT-based support system for dietary counselling, covering two use cases, facilitation and support of the work of nutritionists in ambulatory allergological nutrition counselling and of nutritionists involved in the care of geriatric patients, especially of those with frailty. One of the project's significant features is that the user's sensitive data remain under his or her personal control at all times.

Nowadays, one standard therapy for advanced heart failure is the implantation of left ventricular assist devices (LVAD). The LVAD is a mechanical pump, which supports the heart to pump blood to the rest of the body. However, there are serious complications including pump thrombosis. Pump thrombosis is a formation of blood clot inside the LVAD pump, which can result in pump stop and worst case leads to death. To increase the probability of success in thrombosis treatment, an early diagnosis of pump thrombosis is important. An often mentioned indication for pump thrombosis is an increase in pump current. For this reason, this paper analyzes the log files of LVAD patients. Four algorithms were developed and applied to the measured pump current with the aim to detect an increase in pump current earlier as it is actually done by the physician. The first algorithm incorporated different absolute thresholds. In addition, the second algorithm considered the circadian fluctuation to estimate suitable thresholds. The third and fourth algorithms applied a relative threshold based on the standard deviation of the pump current. The third algorithm applied a standard deviation, which is calculated for each patient and each hours of the day individually. The fourth algorithm used for each patient an individual average standard deviation. This paper shows the results of 5 pump thromboses curves analyzed with the different algorithms each. Using the fourth algorithm, a pump current increase was detected approximately 4 days earlier compared to the “High Watt” alarm caused by the LVAD controller.

In this paper, results from the long-term usage of a mobile application (app) for seniors that encourages physical and mental activity are presented. The application was designed for elderly inhabitants of senior residences to motivate them to increase their physical and mental activity in everyday life. Usage statistics of 82 users for about two years were processed and show that the active elderly users can be clustered in two groups with either increasing or decreasing and very little constant activity. Users with decreasing activity have also shown decreasing usage errors with the app's user interface which may indicate that they are growing out of the app. The results show insight view about the usage and suggest that the Concept of Flow can be applied here.

Assessing a person's physical activity is an important challenge in health studies. However, questionnaires are often inaccurate, precise monitoring devices such as high-end activity monitors are expensive and obtrusive to use, and more easily available measures such as step count are often too simple for meaningful conclusions. We therefore investigated whether data available from modern low-cost activity trackers can be used to estimate calories burnt in physical activity for cardiac prevention. We collected ground truth data from 8 users covering 157 activities on 50 days. We implemented a threshold approach for step data analysis and compared the results with the user's annotations. Compared to the commonly used IPAQ questionnaire our approach is more reliable. Our result indicates that low-cost activity trackers are a feasible tool for assessing physical activity in certain settings.

We present a new control approach for a wearable artificial kidney device, which is currently under development. In contrast to conventional dialysis, this device makes use of a selective adsorption filter which enables miniaturization and thereby a mobile application also out of clinical settings. To facilitate such an unsupervised treatment the device is equipped with diverse sensing capabilities to adapt its function and so allows to maintain the hydration level as well as different concentrations in the blood. This paper elaborates on the developed combined control approach for these interacting main functions which allow the personalization of the renal support system.

This paper presents a novel approach to fully automate the Timed Up and Go Assessment Test (TUG) in professional environments. The approach, called Skeleton Timed Up and Go (sTUG), is based on the usage of two Kinect for Xbox 360 sensors. sTUG supports the execution and documentation of the traditional TUG assessment test and furthermore calculates nine events, which demarcate the five main components during a run. On two days we conducted an experiment with five elderly aged 70-84 and four males aged 29-31 in the activity laboratory of the OFFIS Institute, Oldenburg to proof the reliability and feasibility of the system. Results demonstrate that sTUG can precisely measure the total duration of traditional TUG and is capable of detecting accurately nine motion events which demarcate the components during a run.

Conventional dialysis therapy ofpatients with end-stage renal disease is time-consuming and its discontinuous nature induces physiological stress to the patient. Therefore new emerging approaches facilitate prolonged treatment by using portable devices in home-settings. Such unsupervised treatment calls for continuous automated monitoring of various physiological parameters. This paper highlights the challenges of autonomous personalized renal treatment and presents a leap towards contextaware monitoring in data streams by a physiological modelbased approach. The algorithmic principle is employed to propagate urea concentration, which is measured to assess dialysis efficacy. Simulation results validate the algorithm for the application in a mobile dialysis therapy currently under development. The presented method enables the estimation of medically significant, but technically inaccessible physiological parameters from data stream propagation.

The demand for unattended aerial systems capable of fulfilling e.g. surveillance tasks in contaminated or inaccessible areas without any assistance of a human pilot is the motivation for the investigation of a 3D trajectory control. Hence, this paper deals with the development of such a control algorithm able to follow any kind of 3D trajectory within the quadrocopter's capabilities. In this paper the 3D trajectory control algorithm is described. Many simulations were done to find the optimal trajectory course. Further, the control algorithm was implemented on a flight demonstrator for validation and experimental results regarding linear and circle trajectories are provided.

Motivation for the investigation of position and waypoint controllers is the demand for Unattended Aerial Systems (UAS) capable of fulfilling e.g. surveillance tasks in contaminated or in inaccessible areas. Hence, this paper deals with the development of a 2D GPS-based position control system for 4 Rotor Helicopters able to keep positions above given destinations as well as to navigate between waypoints while minimizing trajectory errors. Additionally, the novel control system enables permanent full speed flight with reliable altitude keeping considering that the resulting lift is decreasing while changing pitch or roll angles for position control. In the following chapters the control procedure for position control and waypoint navigation is described. The dynamic behavior was simulated by means of Matlab/Simulink and results are shown. Further, the control strategies were implemented on a flight demonstrator for validation, experimental results are provided and a comparison is discussed.

As a response to the challenge of providing high-quality healthcare services with reasonable costs while the elderly population and the associated chronic diseases increase, SAPHIRE architecture provides an intelligent healthcare monitoring architecture. The monitoring of patients is achieved through a clinical decision support system based on clinical guidelines. SAPHIRE provides the necessary interoperability layers to access the patient's vital signs from wireless medical sensors and the electronic healthcare records of the patient in order to exploit them in the decision process seamlessly. This architecture is presented through two pilot applications: one for the bedside monitoring of cardiac patients at hospitals, and the other for homecare monitoring of the cardiac patients rehabilitated after a revascularisation therapy.

In this paper the concept and first implementation of a robotic system for assistance during the milling in middle ear surgery is presented. Novel aspects of this system are the new force sensing approach, a parallel kinematic structure and the modified zero-force control scheme. In contrary to other force sensors the proposed approach only requires the position data delivered by a 6 d.o.f. position/orientation measurement system and properties of elastic objects. Because 6 d.o.f. position/orientation measurement systems are standard equipment in computer-aided surgery there is no need for additional sensors. The paper concludes with the experimental evaluation of the force control scheme and an accuracy analysis for the assisted milling

In this paper we describe a new system for the navigated viewing of medical image data and the different visualization concepts being realized within the system and discuss the usefulness of the different visualization modes. The system comprises a compact and mobile viewing device that can be tracked by an optical position sensor. It is capable of displaying reasonable parts of a medical image data set (i.e. CT) on its screen in relation to the spatial position regarding a patient. Requirements of simple integration in a clinical environment and various visualization modes, to be applied in maxillofacial surgery are described.

In this paper a robotic system for the application in head and spinal surgery is described. The system, developed as the Surgical Robotics Lab (SRL), is distinguished from other systems by the complete interactive control by the surgeon. From the viewpoint of telemanipulation it can be interpreted as a combined master/slave manipulator. This system has already been applied clinically in maxillofacial surgery for the exact placement of implants. In this paper extensions are described for the use in spinal surgery. The advantages and arising problems are described, especially for shaping at vertebrae for the decompression of the nerve channel.

Describes a method to detect the contact position and time/event for force/torque guided robots. The method is of special importance for interactively used robots, because the simple observation of the z-component of the force and its derivative is not sufficient in surgical environments. In these environments an undesired contact of a surgeon with the robot or an excessive change of the guiding force can be misinterpreted by the controller. For surgical applications it is important, that the surgeon can handle the system easily and intuitively. Consequently, in the described approach the system can be controlled by the surgeon by grasping the surgical tool (drilling machine, shaper, or saw) mounted on the manipulator and moving it as usual. In the paper, control schemes are proposed to achieve these features of a robotics system while the surgeon applies at the same time force to the TCP.