Nonlinear analysis of actigraphic signals for the assessment of the attention-deficit/hyperactivity disorder (ADHD)

Attention-deficit/hyperactivity disorder (ADHD) is the most common neurobehavioral disorder in children and adolescents; however, its etiology is still unknown, which hinders the existence of reliable, fast and inexpensive standard diagnostic methods. In this paper, we propose a novel methodology for automatic diagnosis of the combined type of ADHD based on nonlinear signal processing of 24h-long actigraphic registries. Since it relies on actigraphy measurements, it constitutes an inexpensive and non-invasive objective diagnostic method. Our results on real data reach 96.77% sensitivity and 84.38% specificity by means of multidimensional classifiers driven by combined features from different time intervals. Our analysis also reveals that, if features from a single time interval are used, the whole 24-h interval is the only one that yields classification figures with practical diagnostic capabilities. Overall, our figures overcome those obtained by actigraphy-based methods reported and are comparable with others based on more expensive (and not so convenient) adquisition methods.

Automatic detection of wakefulness and rest intervals in actigraphic signals: a data-driven approach

Actigraphy is an useful tool for evaluating the activity pattern of a subject; activity registries are usually processed by first splitting the signal into its wakefulness and rest intervals and then analyzing each one in isolation. Consequently, a preprocessing stage for such a splitting is needed. Several methods have been reported to this end but they rely on parameters and thresholds which are manually set based on previous knowledge of the signals or learned from training. This compromises the general applicability of this methods. In this paper we propose a new method in which thresholds are automatically set based solely on the specific registry to be analyzed. The method consists of two stages: (1) estimation of an initial classification mask by means of the expectation maximization algorithm and (2) estimation of a final refined mask through an iterative method which re-estimates both the mask and the classifier parameters at each iteration step. Results on real data show that our methodology outperforms those so far proposed and can be more effectively used to obtain derived sleep quality parameters from actigraphy registries.