Validation of a Smartphone-Based Approach to In Situ Cognitive Fatigue Assessment.

BACKGROUND: Acquired Brain Injuries (ABIs) can result in multiple detrimental cognitive effects, such as reduced memory capability, concentration, and planning. These effects can lead to cognitive fatigue, which can exacerbate the symptoms of ABIs and hinder management and recovery. Assessing cognitive fatigue is difficult due to the largely subjective nature of the condition and existing assessment approaches. Traditional methods of assessment use self-assessment questionnaires delivered in a medical setting, but recent work has attempted to employ more objective cognitive tests as a way of evaluating cognitive fatigue. However, these tests are still predominantly delivered within a medical environment, limiting their utility and efficacy. OBJECTIVE: The aim of this research was to investigate how cognitive fatigue can be accurately assessed in situ, during the quotidian activities of life. It was hypothesized that this assessment could be achieved through the use of mobile assistive technology to assess working memory, sustained attention, information processing speed, reaction time, and cognitive throughput. METHODS: The study used a bespoke smartphone app to track daily cognitive performance, in order to assess potential levels of cognitive fatigue. Twenty-one participants with no prior reported brain injuries took place in a two-week study, resulting in 81 individual testing instances being collected. The smartphone app delivered three cognitive tests on a daily basis: (1) Spatial Span to measure visuospatial working memory; (2) Psychomotor Vigilance Task (PVT) to measure sustained attention, information processing speed, and reaction time; and (3) a Mental Arithmetic Test to measure cognitive throughput. A smartphone-optimized version of the Mental Fatigue Scale (MFS) self-assessment questionnaire was used as a baseline to assess the validity of the three cognitive tests, as the questionnaire has already been validated in multiple peer-reviewed studies. RESULTS: The most highly correlated results were from the PVT, which showed a positive correlation with those from the prevalidated MFS, measuring 0.342 (P<.008). Scores from the cognitive tests were entered into a regression model and showed that only reaction time in the PVT was a significant predictor of fatigue (P=.016, F=2.682, 95% CI 9.0-84.2). Higher scores on the MFS were related to increases in reaction time during our mobile variant of the PVT. CONCLUSIONS: The results show that the PVT mobile cognitive test developed for this study could be used as a valid and reliable method for measuring cognitive fatigue in situ. This test would remove the subjectivity associated with established self-assessment approaches and the need for assessments to be performed in a medical setting. Based on our findings, future work could explore delivering a small set of tests with increased duration to further improve measurement reliability. Moreover, as the smartphone assessment tool can be used as part of everyday life, additional sources of data relating to physiological, psychological, and environmental context could be included within the analysis to improve the nature and precision of the assessment process.

A predictive model for assistive technology adoption for people with dementia.

Assistive technology has the potential to enhance the level of independence of people with dementia, thereby increasing the possibility of supporting home-based care. In general, people with dementia are reluctant to change; therefore, it is important that suitable assistive technologies are selected for them. Consequently, the development of predictive models that are able to determine a person's potential to adopt a particular technology is desirable. In this paper, a predictive adoption model for a mobile phone-based video streaming system, developed for people with dementia, is presented. Taking into consideration characteristics related to a person's ability, living arrangements, and preferences, this paper discusses the development of predictive models, which were based on a number of carefully selected data mining algorithms for classification. For each, the learning on different relevant features for technology adoption has been tested, in conjunction with handling the imbalance of available data for output classes. Given our focus on providing predictive tools that could be used and interpreted by healthcare professionals, models with ease-of-use, intuitive understanding, and clear decision making processes are preferred. Predictive models have, therefore, been evaluated on a multi-criterion basis: in terms of their prediction performance, robustness, bias with regard to two types of errors and usability. Overall, the model derived from incorporating a k-Nearest-Neighbour algorithm using seven features was found to be the optimal classifier of assistive technology adoption for people with dementia (prediction accuracy 0.84 ± 0.0242).

Addressing the challenges of sleep/wake class imbalance in bed based non-contact actigraphic recordings of sleep.

Utilising strategically positioned bed-mounted accelerometers, the Passive Sleep Actigraphy platform aims to deliver a non-contact method for identifying periods of wakefulness during night-time sleep. One of the key problems in developing data driven approaches for automatic sleep monitoring is managing the inherent sleep/wake class imbalance. In the current study, actigraphy data from three participants over a period of 30 days was collected. Upon examination, it was found that only 10% contained wake data. Consequently, this resulted in classifier overfitting to the majority class (sleep), thereby impeding the ability of the Passive Sleep Actigraphy platform to correctly identify periods of wakefulness during sleep; a key measure in the identification of sleep problems. Utilising Spread Subsample and Synthetic Minority Oversampling Techniques, this paper demonstrates a potential solution to this issue, reporting improvements of up to 28% in wake detection when compared to baseline data while maintaining an overall classifier accuracy of 90%.