Filtern
Dokumenttyp
- Wissenschaftlicher Artikel (10)
- Postprint (6)
- Monographie/Sammelband (2)
- Bericht (1)
Gehört zur Bibliographie
- ja (19)
Schlagworte
- machine learning (6)
- inertial measurement unit (4)
- OptoGait (2)
- RGB-D cameras (2)
- X-ray (2)
- Zebris (2)
- comparison of devices (2)
- data pipeline (2)
- digital health (2)
- disorder recognition (2)
Institut
- Hasso-Plattner-Institut für Digital Engineering GmbH (19) (entfernen)
Privacy regulations and the physical distribution of heterogeneous data are often primary concerns for the development of deep learning models in a medical context. This paper evaluates the feasibility of differentially private federated learning for chest X-ray classification as a defense against data privacy attacks. To the best of our knowledge, we are the first to directly compare the impact of differentially private training on two different neural network architectures, DenseNet121 and ResNet50. Extending the federated learning environments previously analyzed in terms of privacy, we simulated a heterogeneous and imbalanced federated setting by distributing images from the public CheXpert and Mendeley chest X-ray datasets unevenly among 36 clients. Both non-private baseline models achieved an area under the receiver operating characteristic curve (AUC) of 0.940.94 on the binary classification task of detecting the presence of a medical finding. We demonstrate that both model architectures are vulnerable to privacy violation by applying image reconstruction attacks to local model updates from individual clients. The attack was particularly successful during later training stages. To mitigate the risk of a privacy breach, we integrated Rényi differential privacy with a Gaussian noise mechanism into local model training. We evaluate model performance and attack vulnerability for privacy budgets ε∈{1,3,6,10}�∈{1,3,6,10}. The DenseNet121 achieved the best utility-privacy trade-off with an AUC of 0.940.94 for ε=6�=6. Model performance deteriorated slightly for individual clients compared to the non-private baseline. The ResNet50 only reached an AUC of 0.760.76 in the same privacy setting. Its performance was inferior to that of the DenseNet121 for all considered privacy constraints, suggesting that the DenseNet121 architecture is more robust to differentially private training.
Privacy regulations and the physical distribution of heterogeneous data are often primary concerns for the development of deep learning models in a medical context. This paper evaluates the feasibility of differentially private federated learning for chest X-ray classification as a defense against data privacy attacks. To the best of our knowledge, we are the first to directly compare the impact of differentially private training on two different neural network architectures, DenseNet121 and ResNet50. Extending the federated learning environments previously analyzed in terms of privacy, we simulated a heterogeneous and imbalanced federated setting by distributing images from the public CheXpert and Mendeley chest X-ray datasets unevenly among 36 clients. Both non-private baseline models achieved an area under the receiver operating characteristic curve (AUC) of 0.940.94 on the binary classification task of detecting the presence of a medical finding. We demonstrate that both model architectures are vulnerable to privacy violation by applying image reconstruction attacks to local model updates from individual clients. The attack was particularly successful during later training stages. To mitigate the risk of a privacy breach, we integrated Rényi differential privacy with a Gaussian noise mechanism into local model training. We evaluate model performance and attack vulnerability for privacy budgets ε∈{1,3,6,10}�∈{1,3,6,10}. The DenseNet121 achieved the best utility-privacy trade-off with an AUC of 0.940.94 for ε=6�=6. Model performance deteriorated slightly for individual clients compared to the non-private baseline. The ResNet50 only reached an AUC of 0.760.76 in the same privacy setting. Its performance was inferior to that of the DenseNet121 for all considered privacy constraints, suggesting that the DenseNet121 architecture is more robust to differentially private training.
How We Found Our IMU
(2020)
Inertial measurement units (IMUs) are commonly used for localization or movement tracking in pervasive healthcare-related studies, and gait analysis is one of the most often studied topics using IMUs. The increasing variety of commercially available IMU devices offers convenience by combining the sensor modalities and simplifies the data collection procedures. However, selecting the most suitable IMU device for a certain use case is increasingly challenging. In this study, guidelines for IMU selection are proposed. In particular, seven IMUs were compared in terms of their specifications, data collection procedures, and raw data quality. Data collected from the IMUs were then analyzed by a gait analysis algorithm. The difference in accuracy of the calculated gait parameters between the IMUs could be used to retrace the issues in raw data, such as acceleration range or sensor calibration. Based on our algorithm, we were able to identify the best-suited IMUs for our needs. This study provides an overview of how to select the IMUs based on the area of study with concrete examples, and gives insights into the features of seven commercial IMUs using real data.
How We Found Our IMU
(2020)
Inertial measurement units (IMUs) are commonly used for localization or movement tracking in pervasive healthcare-related studies, and gait analysis is one of the most often studied topics using IMUs. The increasing variety of commercially available IMU devices offers convenience by combining the sensor modalities and simplifies the data collection procedures. However, selecting the most suitable IMU device for a certain use case is increasingly challenging. In this study, guidelines for IMU selection are proposed. In particular, seven IMUs were compared in terms of their specifications, data collection procedures, and raw data quality. Data collected from the IMUs were then analyzed by a gait analysis algorithm. The difference in accuracy of the calculated gait parameters between the IMUs could be used to retrace the issues in raw data, such as acceleration range or sensor calibration. Based on our algorithm, we were able to identify the best-suited IMUs for our needs. This study provides an overview of how to select the IMUs based on the area of study with concrete examples, and gives insights into the features of seven commercial IMUs using real data.
Inertial measurement units (IMUs) enable easy to operate and low-cost data recording for gait analysis. When combined with treadmill walking, a large number of steps can be collected in a controlled environment without the need of a dedicated gait analysis laboratory. In order to evaluate existing and novel IMU-based gait analysis algorithms for treadmill walking, a reference dataset that includes IMU data as well as reliable ground truth measurements for multiple participants and walking speeds is needed. This article provides a reference dataset consisting of 15 healthy young adults who walked on a treadmill at three different speeds. Data were acquired using seven IMUs placed on the lower body, two different reference systems (Zebris FDMT-HQ and OptoGait), and two RGB cameras. Additionally, in order to validate an existing IMU-based gait analysis algorithm using the dataset, an adaptable modular data analysis pipeline was built. Our results show agreement between the pressure-sensitive Zebris and the photoelectric OptoGait system (r = 0.99), demonstrating the quality of our reference data. As a use case, the performance of an algorithm originally designed for overground walking was tested on treadmill data using the data pipeline. The accuracy of stride length and stride time estimations was comparable to that reported in other studies with overground data, indicating that the algorithm is equally applicable to treadmill data. The Python source code of the data pipeline is publicly available, and the dataset will be provided by the authors upon request, enabling future evaluations of IMU gait analysis algorithms without the need of recording new data.
TRIPOD
(2021)
Inertial measurement units (IMUs) enable easy to operate and low-cost data recording for gait analysis. When combined with treadmill walking, a large number of steps can be collected in a controlled environment without the need of a dedicated gait analysis laboratory. In order to evaluate existing and novel IMU-based gait analysis algorithms for treadmill walking, a reference dataset that includes IMU data as well as reliable ground truth measurements for multiple participants and walking speeds is needed. This article provides a reference dataset consisting of 15 healthy young adults who walked on a treadmill at three different speeds. Data were acquired using seven IMUs placed on the lower body, two different reference systems (Zebris FDMT-HQ and OptoGait), and two RGB cameras. Additionally, in order to validate an existing IMU-based gait analysis algorithm using the dataset, an adaptable modular data analysis pipeline was built. Our results show agreement between the pressure-sensitive Zebris and the photoelectric OptoGait system (r = 0.99), demonstrating the quality of our reference data. As a use case, the performance of an algorithm originally designed for overground walking was tested on treadmill data using the data pipeline. The accuracy of stride length and stride time estimations was comparable to that reported in other studies with overground data, indicating that the algorithm is equally applicable to treadmill data. The Python source code of the data pipeline is publicly available, and the dataset will be provided by the authors upon request, enabling future evaluations of IMU gait analysis algorithms without the need of recording new data.
Giving emotional intelligence to machines can facilitate the early detection and prediction of mental diseases and symptoms. Electroencephalography (EEG)-based emotion recognition is widely applied because it measures electrical correlates directly from the brain rather than indirect measurement of other physiological responses initiated by the brain. Therefore, we used non-invasive and portable EEG sensors to develop a real-time emotion classification pipeline. The pipeline trains different binary classifiers for Valence and Arousal dimensions from an incoming EEG data stream achieving a 23.9% (Arousal) and 25.8% (Valence) higher F1-Score on the state-of-art AMIGOS dataset than previous work. Afterward, the pipeline was applied to the curated dataset from 15 participants using two consumer-grade EEG devices while watching 16 short emotional videos in a controlled environment. Mean F1-Scores of 87% (Arousal) and 82% (Valence) were achieved for an immediate label setting. Additionally, the pipeline proved to be fast enough to achieve predictions in real-time in a live scenario with delayed labels while continuously being updated. The significant discrepancy from the readily available labels on the classification scores leads to future work to include more data. Thereafter, the pipeline is ready to be used for real-time applications of emotion classification.
Proceedings of the HPI Research School on Service-oriented Systems Engineering 2020 Fall Retreat
(2021)
Design and Implementation of service-oriented architectures imposes a huge number of research questions from the fields of software engineering, system analysis and modeling, adaptability, and application integration. Component orientation and web services are two approaches for design and realization of complex web-based system. Both approaches allow for dynamic application adaptation as well as integration of enterprise application.
Service-Oriented Systems Engineering represents a symbiosis of best practices in object-orientation, component-based development, distributed computing, and business process management. It provides integration of business and IT concerns.
The annual Ph.D. Retreat of the Research School provides each member the opportunity to present his/her current state of their research and to give an outline of a prospective Ph.D. thesis. Due to the interdisciplinary structure of the research school, this technical report covers a wide range of topics. These include but are not limited to: Human Computer Interaction and Computer Vision as Service; Service-oriented Geovisualization Systems; Algorithm Engineering for Service-oriented Systems; Modeling and Verification of Self-adaptive Service-oriented Systems; Tools and Methods for Software Engineering in Service-oriented Systems; Security Engineering of Service-based IT Systems; Service-oriented Information Systems; Evolutionary Transition of Enterprise Applications to Service Orientation; Operating System Abstractions for Service-oriented Computing; and Services Specification, Composition, and Enactment.
The “HPI Future SOC Lab” is a cooperation of the Hasso Plattner Institute (HPI) and industry partners. Its mission is to enable and promote exchange and interaction between the research community and the industry partners.
The HPI Future SOC Lab provides researchers with free of charge access to a complete infrastructure of state of the art hard and software. This infrastructure includes components, which might be too expensive for an ordinary research environment, such as servers with up to 64 cores and 2 TB main memory. The offerings address researchers particularly from but not limited to the areas of computer science and business information systems. Main areas of research include cloud computing, parallelization, and In-Memory technologies.
This technical report presents results of research projects executed in 2019. Selected projects have presented their results on April 9th and November 12th 2019 at the Future SOC Lab Day events.
Recent trends in ubiquitous computing have led to a proliferation of studies that focus on human activity recognition (HAR) utilizing inertial sensor data that consist of acceleration, orientation and angular velocity. However, the performances of such approaches are limited by the amount of annotated training data, especially in fields where annotating data is highly time-consuming and requires specialized professionals, such as in healthcare. In image classification, this limitation has been mitigated by powerful oversampling techniques such as data augmentation. Using this technique, this work evaluates to what extent transforming inertial sensor data into movement trajectories and into 2D heatmap images can be advantageous for HAR when data are scarce. A convolutional long short-term memory (ConvLSTM) network that incorporates spatiotemporal correlations was used to classify the heatmap images. Evaluation was carried out on Deep Inertial Poser (DIP), a known dataset composed of inertial sensor data. The results obtained suggest that for datasets with large numbers of subjects, using state-of-the-art methods remains the best alternative. However, a performance advantage was achieved for small datasets, which is usually the case in healthcare. Moreover, movement trajectories provide a visual representation of human activities, which can help researchers to better interpret and analyze motion patterns.
Recent trends in ubiquitous computing have led to a proliferation of studies that focus on human activity recognition (HAR) utilizing inertial sensor data that consist of acceleration, orientation and angular velocity. However, the performances of such approaches are limited by the amount of annotated training data, especially in fields where annotating data is highly time-consuming and requires specialized professionals, such as in healthcare. In image classification, this limitation has been mitigated by powerful oversampling techniques such as data augmentation. Using this technique, this work evaluates to what extent transforming inertial sensor data into movement trajectories and into 2D heatmap images can be advantageous for HAR when data are scarce. A convolutional long short-term memory (ConvLSTM) network that incorporates spatiotemporal correlations was used to classify the heatmap images. Evaluation was carried out on Deep Inertial Poser (DIP), a known dataset composed of inertial sensor data. The results obtained suggest that for datasets with large numbers of subjects, using state-of-the-art methods remains the best alternative. However, a performance advantage was achieved for small datasets, which is usually the case in healthcare. Moreover, movement trajectories provide a visual representation of human activities, which can help researchers to better interpret and analyze motion patterns.
SONAR
(2023)
Accurate and comprehensive nursing documentation is essential to ensure quality patient care. To streamline this process, we present SONAR, a publicly available dataset of nursing activities recorded using inertial sensors in a nursing home. The dataset includes 14 sensor streams, such as acceleration and angular velocity, and 23 activities recorded by 14 caregivers using five sensors for 61.7 hours. The caregivers wore the sensors as they performed their daily tasks, allowing for continuous monitoring of their activities. We additionally provide machine learning models that recognize the nursing activities given the sensor data. In particular, we present benchmarks for three deep learning model architectures and evaluate their performance using different metrics and sensor locations. Our dataset, which can be used for research on sensor-based human activity recognition in real-world settings, has the potential to improve nursing care by providing valuable insights that can identify areas for improvement, facilitate accurate documentation, and tailor care to specific patient conditions.
Psychology and nutritional science research has highlighted the impact of negative emotions and cognitive load on calorie consumption behaviour using subjective questionnaires. Isolated studies in other domains objectively assess cognitive load without considering its effects on eating behaviour. This study aims to explore the potential for developing an integrated eating behaviour assistant system that incorporates cognitive load factors. Two experimental sessions were conducted using custom-developed experimentation software to induce different stimuli. During these sessions, we collected 30 h of physiological, food consumption, and affective states questionnaires data to automatically detect cognitive load and analyse its effect on food choice. Utilising grid search optimisation and leave-one-subject-out cross-validation, a support vector machine model achieved a mean classification accuracy of 85.12% for the two cognitive load tasks using eight relevant features. Statistical analysis was performed on calorie consumption and questionnaire data. Furthermore, 75% of the subjects with higher negative affect significantly increased consumption of specific foods after high-cognitive-load tasks. These findings offer insights into the intricate relationship between cognitive load, affective states, and food choice, paving the way for an eating behaviour assistant system to manage food choices during cognitive load. Future research should enhance system capabilities and explore real-world applications.
Quantifying neurological disorders from voice is a rapidly growing field of research and holds promise for unobtrusive and large-scale disorder monitoring. The data recording setup and data analysis pipelines are both crucial aspects to effectively obtain relevant information from participants. Therefore, we performed a systematic review to provide a high-level overview of practices across various neurological disorders and highlight emerging trends. PRISMA-based literature searches were conducted through PubMed, Web of Science, and IEEE Xplore to identify publications in which original (i.e., newly recorded) datasets were collected. Disorders of interest were psychiatric as well as neurodegenerative disorders, such as bipolar disorder, depression, and stress, as well as amyotrophic lateral sclerosis amyotrophic lateral sclerosis, Alzheimer's, and Parkinson's disease, and speech impairments (aphasia, dysarthria, and dysphonia). Of the 43 retrieved studies, Parkinson's disease is represented most prominently with 19 discovered datasets. Free speech and read speech tasks are most commonly used across disorders. Besides popular feature extraction toolkits, many studies utilise custom-built feature sets. Correlations of acoustic features with psychiatric and neurodegenerative disorders are presented. In terms of analysis, statistical analysis for significance of individual features is commonly used, as well as predictive modeling approaches, especially with support vector machines and a small number of artificial neural networks. An emerging trend and recommendation for future studies is to collect data in everyday life to facilitate longitudinal data collection and to capture the behavior of participants more naturally. Another emerging trend is to record additional modalities to voice, which can potentially increase analytical performance.
Quantifying neurological disorders from voice is a rapidly growing field of research and holds promise for unobtrusive and large-scale disorder monitoring. The data recording setup and data analysis pipelines are both crucial aspects to effectively obtain relevant information from participants. Therefore, we performed a systematic review to provide a high-level overview of practices across various neurological disorders and highlight emerging trends. PRISMA-based literature searches were conducted through PubMed, Web of Science, and IEEE Xplore to identify publications in which original (i.e., newly recorded) datasets were collected. Disorders of interest were psychiatric as well as neurodegenerative disorders, such as bipolar disorder, depression, and stress, as well as amyotrophic lateral sclerosis amyotrophic lateral sclerosis, Alzheimer's, and Parkinson's disease, and speech impairments (aphasia, dysarthria, and dysphonia). Of the 43 retrieved studies, Parkinson's disease is represented most prominently with 19 discovered datasets. Free speech and read speech tasks are most commonly used across disorders. Besides popular feature extraction toolkits, many studies utilise custom-built feature sets. Correlations of acoustic features with psychiatric and neurodegenerative disorders are presented. In terms of analysis, statistical analysis for significance of individual features is commonly used, as well as predictive modeling approaches, especially with support vector machines and a small number of artificial neural networks. An emerging trend and recommendation for future studies is to collect data in everyday life to facilitate longitudinal data collection and to capture the behavior of participants more naturally. Another emerging trend is to record additional modalities to voice, which can potentially increase analytical performance.
Background
The aggregation of a series of N-of-1 trials presents an innovative and efficient study design, as an alternative to traditional randomized clinical trials. Challenges for the statistical analysis arise when there is carry-over or complex dependencies of the treatment effect of interest.
Methods
In this study, we evaluate and compare methods for the analysis of aggregated N-of-1 trials in different scenarios with carry-over and complex dependencies of treatment effects on covariates. For this, we simulate data of a series of N-of-1 trials for Chronic Nonspecific Low Back Pain based on assumed causal relationships parameterized by directed acyclic graphs. In addition to existing statistical methods such as regression models, Bayesian Networks, and G-estimation, we introduce a carry-over adjusted parametric model (COAPM).
Results
The results show that all evaluated existing models have a good performance when there is no carry-over and no treatment dependence. When there is carry-over, COAPM yields unbiased and more efficient estimates while all other methods show some bias in the estimation. When there is known treatment dependence, all approaches that are capable to model it yield unbiased estimates. Finally, the efficiency of all methods decreases slightly when there are missing values, and the bias in the estimates can also increase.
Conclusions
This study presents a systematic evaluation of existing and novel approaches for the statistical analysis of a series of N-of-1 trials. We derive practical recommendations which methods may be best in which scenarios.
Gait analysis is an important tool for the early detection of neurological diseases and for the assessment of risk of falling in elderly people. The availability of low-cost camera hardware on the market today and recent advances in Machine Learning enable a wide range of clinical and health-related applications, such as patient monitoring or exercise recognition at home. In this study, we evaluated the motion tracking performance of the latest generation of the Microsoft Kinect camera, Azure Kinect, compared to its predecessor Kinect v2 in terms of treadmill walking using a gold standard Vicon multi-camera motion capturing system and the 39 marker Plug-in Gait model. Five young and healthy subjects walked on a treadmill at three different velocities while data were recorded simultaneously with all three camera systems. An easy-to-administer camera calibration method developed here was used to spatially align the 3D skeleton data from both Kinect cameras and the Vicon system. With this calibration, the spatial agreement of joint positions between the two Kinect cameras and the reference system was evaluated. In addition, we compared the accuracy of certain spatio-temporal gait parameters, i.e., step length, step time, step width, and stride time calculated from the Kinect data, with the gold standard system. Our results showed that the improved hardware and the motion tracking algorithm of the Azure Kinect camera led to a significantly higher accuracy of the spatial gait parameters than the predecessor Kinect v2, while no significant differences were found between the temporal parameters. Furthermore, we explain in detail how this experimental setup could be used to continuously monitor the progress during gait rehabilitation in older people.
Gait analysis is an important tool for the early detection of neurological diseases and for the assessment of risk of falling in elderly people. The availability of low-cost camera hardware on the market today and recent advances in Machine Learning enable a wide range of clinical and health-related applications, such as patient monitoring or exercise recognition at home. In this study, we evaluated the motion tracking performance of the latest generation of the Microsoft Kinect camera, Azure Kinect, compared to its predecessor Kinect v2 in terms of treadmill walking using a gold standard Vicon multi-camera motion capturing system and the 39 marker Plug-in Gait model. Five young and healthy subjects walked on a treadmill at three different velocities while data were recorded simultaneously with all three camera systems. An easy-to-administer camera calibration method developed here was used to spatially align the 3D skeleton data from both Kinect cameras and the Vicon system. With this calibration, the spatial agreement of joint positions between the two Kinect cameras and the reference system was evaluated. In addition, we compared the accuracy of certain spatio-temporal gait parameters, i.e., step length, step time, step width, and stride time calculated from the Kinect data, with the gold standard system. Our results showed that the improved hardware and the motion tracking algorithm of the Azure Kinect camera led to a significantly higher accuracy of the spatial gait parameters than the predecessor Kinect v2, while no significant differences were found between the temporal parameters. Furthermore, we explain in detail how this experimental setup could be used to continuously monitor the progress during gait rehabilitation in older people.