Indirectly-Supervised Anomaly Detection of Clinically-Meaningful Health Events from Smart Home Data.

Anomaly detection techniques can extract a wealth of information about unusual events. Unfortunately, these methods yield an abundance of findings that are not of interest, obscuring relevant anomalies. In this work, we improve upon traditional anomaly detection methods by introducing Isudra, an Indirectly-Supervised Detector of Relevant Anomalies from time series data. Isudra employs Bayesian optimization to select time scales, features, base detector algorithms, and algorithm hyperparameters that increase true positive and decrease false positive detection. This optimization is driven by a small amount of example anomalies, driving an indirectly-supervised approach to anomaly detection. Additionally, we enhance the approach by introducing a warm start method that reduces optimization time between similar problems. We validate the feasibility of Isudra to detect clinically-relevant behavior anomalies from over 2 million sensor readings collected in 5 smart homes, reflecting 26 health events. Results indicate that indirectly-supervised anomaly detection outperforms both supervised and unsupervised algorithms at detecting instances of health-related anomalies such as falls, nocturia, depression, and weakness.

The development of a manual-based digital memory notebook intervention with case study illustrations.

There is currently a need to identify feasible and effective interventions to help older individuals suffering from memory loss maintain functional independence and quality of life. To improve upon paper and pencil memory notebook interventions, the Digital Memory Notebook (DMN) application (app) was developed iteratively with persons with cognitive impairment. In this paper we detail a manual-based intervention for training use of the DMN app. A series of three case studies are described to illustrate the clinical process of the DMN intervention, the key components of the intervention and participants' perceptions of the intervention. The Reliable Change Index was applied to pre/post intervention scores that examined everyday memory lapses, daily functioning, coping self-efficacy, satisfaction with life, and quality of life with standardized measures. Following the intervention, two of three participants self-reported a clinically significant reduction in everyday memory lapses and improved everyday functioning. One participant reported clinically significant change in quality of life. All participants demonstrated clinically significant changes in their ability to cope with problems and build self-efficacy. Furthermore, all participants scored in the normative range post-intervention on the measure of satisfaction with life. Clinical observations and participant feedback were used for refinement of the DMN intervention (ClinicalTrials.gov NCT03453554).

Creating a digital memory notebook application for individuals with mild cognitive impairment to support everyday functioning.

Purpose: Memory impairment can necessitate use of external memory aids to preserve functional independence. As external aids can be difficult to learn and remember to use, technology may improve the efficacy of current rehabilitation strategies. We detail the iterative development of a digital application of a paper-and-pencil memory notebook.Methods: Twenty participants (aged 54+) of varying levels of cognitive ability were recruited for four iterations of usability testing (five participants per iteration). Each participant completed a series of tasks using the digital memory notebook (DMN), followed by questionnaires that assessed satisfaction (Questionnaire for User Interface Satisfaction) and usability ratings (Post-Study System Usability Questionnaire) for the application.Results and Conclusions: Between Iterations 2 and 5, participants demonstrated marked reductions in time to complete several types of tasks (e.g., add event, navigate interface) using the DMN. Participants in Iteration 5 also rated all subscales of both the usability and satisfaction questionnaires very highly. Faster task completion times were correlated with more favourable system ratings. However, neither task performance times nor system ratings were correlated with cognitive abilities, scheduling tool use or comfort with technology. Both the questionnaire and performance-based data indicate the final iteration of the DMN was easy to use. Furthermore, the application was user-friendly despite individual differences in cognitive ability, familiarity with scheduling tools and comfort with technology. Future work will demonstrate whether the DMN will support everyday retrospective and prospective memory lapses and increase the functional independence and quality of life for persons with cognitive impairment.Implications for rehabilitationBuilding on practice standards and user-centred design, the digital memory notebook (DMN) application is an "all-in-one" memory aid and organizational tool with an intuitive interface designed to help improve everyday functioning.The DMN's today page, to do list and calendar functions can support everyday prospective and retrospective memory abilities.The DMN's notes, journaling and motivational functions can support longer-term goal planning and mood management.The DMN's alarm functions can support learning to use the DMN and serve as reminders to support prospective memory and aid in activity completion.

SynSys: A Synthetic Data Generation System for Healthcare Applications.

Creation of realistic synthetic behavior-based sensor data is an important aspect of testing machine learning techniques for healthcare applications. Many of the existing approaches for generating synthetic data are often limited in terms of complexity and realism. We introduce SynSys, a machine learning-based synthetic data generation method, to improve upon these limitations. We use this method to generate synthetic time series data that is composed of nested sequences using hidden Markov models and regression models which are initially trained on real datasets. We test our synthetic data generation technique on a real annotated smart home dataset. We use time series distance measures as a baseline to determine how realistic the generated data is compared to real data and demonstrate that SynSys produces more realistic data in terms of distance compared to random data generation, data from another home, and data from another time period. Finally, we apply our synthetic data generation technique to the problem of generating data when only a small amount of ground truth data is available. Using semi-supervised learning we demonstrate that SynSys is able to improve activity recognition accuracy compared to using the small amount of real data alone.

Smart Secure Homes: A Survey of Smart Home Technologies that Sense, Assess, and Respond to Security Threats.

Smart home design has undergone a metamorphosis in recent years. The field has evolved from designing theoretical smart home frameworks and performing scripted tasks in laboratories. Instead, we now find robust smart home technologies that are commonly used by large segments of the population in a variety of settings. Recent smart home applications are focused on activity recognition, health monitoring, and automation. In this paper, we take a look at another important role for smart homes: security. We first explore the numerous ways smart homes can and do provide protection for their residents. Next, we provide a comparative analysis of the alternative tools and research that has been developed for this purpose. We investigate not only existing commercial products that have been introduced but also discuss the numerous research that has been focused on detecting and identifying potential threats. Finally, we close with open challenges and ideas for future research that will keep individuals secure and healthy while in their own homes.

Activity Learning as a Foundation for Security Monitoring in Smart Homes.

Smart environment technology has matured to the point where it is regularly used in everyday homes as well as research labs. With this maturation of the technology, we can consider using smart homes as a practical mechanism for improving home security. In this paper, we introduce an activity-aware approach to security monitoring and threat detection in smart homes. We describe our approach using the CASAS smart home framework and activity learning algorithms. By monitoring for activity-based anomalies we can detect possible threats and take appropriate action. We evaluate our proposed method using data collected in CASAS smart homes and demonstrate the partnership between activity-aware smart homes and biometric devices in the context of the CASAS on-campus smart apartment testbed.

Multicomponent analysis of a digital Trail Making Test.

OBJECTIVE: The purpose of the current study was to use a newly developed digital tablet-based variant of the TMT to isolate component cognitive processes underlying TMT performance. METHOD: Similar to the paper-based trail making test, this digital variant consists of two conditions, Part A and Part B. However, this digital version automatically collects additional data to create component subtest scores to isolate cognitive abilities. Specifically, in addition to the total time to completion and number of errors, the digital Trail Making Test (dTMT) records several unique components including the number of pauses, pause duration, lifts, lift duration, time inside each circle, and time between circles. Participants were community-dwelling older adults who completed a neuropsychological evaluation including measures of processing speed, inhibitory control, visual working memory/sequencing, and set-switching. The abilities underlying TMT performance were assessed through regression analyses of component scores from the dTMT with traditional neuropsychological measures. RESULTS: Results revealed significant correlations between paper and digital variants of Part A (rs = .541, p < .001) and paper and digital versions of Part B (rs = .799, p < .001). Regression analyses with traditional neuropsychological measures revealed that Part A components were best predicted by speeded processing, while inhibitory control and visual/spatial sequencing were predictors of specific components of Part B. Exploratory analyses revealed that specific dTMT-B components were associated with a performance-based medication management task. CONCLUSIONS: Taken together, these results elucidate specific cognitive abilities underlying TMT performance, as well as the utility of isolating digital components.

An analysis of a digital variant of the Trail Making Test using machine learning techniques.

BACKGROUND: The goal of this work is to develop a digital version of a standard cognitive assessment, the Trail Making Test (TMT), and assess its utility. OBJECTIVE: This paper introduces a novel digital version of the TMT and introduces a machine learning based approach to assess its capabilities. METHODS: Using digital Trail Making Test (dTMT) data collected from (N = 54) older adult participants as feature sets, we use machine learning techniques to analyze the utility of the dTMT and evaluate the insights provided by the digital features. RESULTS: Predicted TMT scores correlate well with clinical digital test scores (r = 0.98) and paper time to completion scores (r = 0.65). Predicted TICS exhibited a small correlation with clinically derived TICS scores (r = 0.12 Part A, r = 0.10 Part B). Predicted FAB scores exhibited a small correlation with clinically derived FAB scores (r = 0.13 Part A, r = 0.29 for Part B). Digitally derived features were also used to predict diagnosis (AUC of 0.65). CONCLUSION: Our findings indicate that the dTMT is capable of measuring the same aspects of cognition as the paper-based TMT. Furthermore, the dTMT's additional data may be able to help monitor other cognitive processes not captured by the paper-based TMT alone.