SciKit Digital Health: Python Package for Streamlined Wearable Inertial Sensor Data Processing.

Wearable inertial sensors are providing enhanced insight into patient mobility and health. Significant research efforts have focused on wearable algorithm design and deployment in both research and clinical settings; however, open-source, general-purpose software tools for processing various activities of daily living are relatively scarce. Furthermore, few studies include code for replication or off-the-shelf software packages. In this work, we introduce SciKit Digital Health (SKDH), a Python software package (Python Software Foundation) containing various algorithms for deriving clinical features of gait, sit to stand, physical activity, and sleep, wrapped in an easily extensible framework. SKDH combines data ingestion, preprocessing, and data analysis methods geared toward modern data science workflows and streamlines the generation of digital endpoints in "good practice" environments by combining all the necessary data processing steps in a single pipeline. Our package simplifies the construction of new data processing pipelines and promotes reproducibility by following a convention over configuration approach, standardizing most settings on physiologically reasonable defaults in healthy adult populations or those with mild impairment. SKDH is open source, as well as free to use and extend under a permissive Massachusetts Institute of Technology license, and is available from GitHub (PfizerRD/scikit-digital-health), the Python Package Index, and the conda-forge channel of Anaconda.

Development of digital measures for nighttime scratch and sleep using wrist-worn wearable devices.

Patients with atopic dermatitis experience increased nocturnal pruritus which leads to scratching and sleep disturbances that significantly contribute to poor quality of life. Objective measurements of nighttime scratching and sleep quantity can help assess the efficacy of an intervention. Wearable sensors can provide novel, objective measures of nighttime scratching and sleep; however, many current approaches were not designed for passive, unsupervised monitoring during daily life. In this work, we present the development and analytical validation of a method that sequentially processes epochs of sample-level accelerometer data from a wrist-worn device to provide continuous digital measures of nighttime scratching and sleep quantity. This approach uses heuristic and machine learning algorithms in a hierarchical paradigm by first determining when the patient intends to sleep, then detecting sleep-wake states along with scratching episodes, and lastly deriving objective measures of both sleep and scratch. Leveraging reference data collected in a sleep laboratory (NCT ID: NCT03490877), results show that sensor-derived measures of total sleep opportunity (TSO; time when patient intends to sleep) and total sleep time (TST) correlate well with reference polysomnography data (TSO: r = 0.72, p < 0.001; TST: r = 0.76, p < 0.001; N = 32). Log transformed sensor derived measures of total scratching duration achieve strong agreement with reference annotated video recordings (r = 0.82, p < 0.001; N = 25). These results support the use of wearable sensors for objective, continuous measurement of nighttime scratching and sleep during daily life.

Target-Specific Action Classification for Automated Assessment of Human Motor Behavior from Video.

Objective monitoring and assessment of human motor behavior can improve the diagnosis and management of several medical conditions. Over the past decade, significant advances have been made in the use of wearable technology for continuously monitoring human motor behavior in free-living conditions. However, wearable technology remains ill-suited for applications which require monitoring and interpretation of complex motor behaviors (e.g., involving interactions with the environment). Recent advances in computer vision and deep learning have opened up new possibilities for extracting information from video recordings. In this paper, we present a hierarchical vision-based behavior phenotyping method for classification of basic human actions in video recordings performed using a single RGB camera. Our method addresses challenges associated with tracking multiple human actors and classification of actions in videos recorded in changing environments with different fields of view. We implement a cascaded pose tracker that uses temporal relationships between detections for short-term tracking and appearance based tracklet fusion for long-term tracking. Furthermore, for action classification, we use pose evolution maps derived from the cascaded pose tracker as low-dimensional and interpretable representations of the movement sequences for training a convolutional neural network. The cascaded pose tracker achieves an average accuracy of 88% in tracking the target human actor in our video recordings, and overall system achieves average test accuracy of 84% for target-specific action classification in untrimmed video recordings.