Database Reference
In-Depth Information
assessment data. They show that the first principal component includes
a diverse set of measures of general intelligence, and appears to be a
good proxy for general neuropsychological integrity, including measures
of intellectual functioning, verbal and nonverbal reasoning, memory, and
complex attention.
Researchers are developing several other techniques for the automatic
detection of cognitive impairments, including automatically observing
users play modified versions of different games. For instance, a mod-
ified version of the game FreeCell [104] is used in many studies. One
study focuses on mouse movement during the game while others focus
on the subject performance over time, comparing it to the performance
of an automated solver. Using the results, it was possible to differenti-
ate the three mildly cognitively impaired subjects from the six others.
Work with several other computer games, specially created to perform
assessments of cognitive impairments is underway with some promising
early results. Researchers have also studied automatically monitoring
mobility because slowed mobility may be a predictor of future cognitive
decline. The time to answer a phone call was used to measure mobility,
as were passive infrared detectors and several models to infer the mo-
bility of subjects more directly as they move about a residence. More
details on these may be obtained from [103].
Mining data from smart environments has also been used for sleep
research [105] on a long-term basis, in a comfortable setting
6
.In-
tial, ambient light, and time data are tracked from a wrist-worn sensor,
and additional night vision footage is used for later expert inspection.
The authors use two different classification techniques to monitor and
classify the night sleep. Classifier 1 use threshold-based segmentation
on a Gaussian model-based classifier that calculates the variance and
mean parameters for the light intensity and motion data from the train-
ing data, and uses a likelihood per minute of the awake state from the
time-use database. Classifier 2 uses HMM-based segmentation to cap-
ture changes in sleep habits and state, and differentiate awake state
from sleep state. The authors have shown that these techniques can
be used for accurate sleep studies while minimizing the intrusiveness of
the sensing environment for patients suffering from sleep disorders and
psychiatric illnesses.
6
The golden standard for observing sleep-wake patterns is polysomnography (PSG) that
captures relevant sleep information with typically 20, mostly wired sensors attached to the
face, torso and limbs of the patient, making it costly, uncomfortable and less feasible over
longer periods.
