Sleep state classification
This project uses the data from Collaborative Home Infant Monitoring Evaluation (CHIME) dataset. Sleep-wake patterns and sleep states are valuable predictors for detection of life-threatening events in infants. This project is based on an attempt to recognize various sleep states in the data recorded by CHIME home monitor. So far activity data captured by an accelerometrer located on a diaper, heart rate variabilty and respiratory variability have been used to classify sleep states. We are also looking into inreasing reliability of sleep state classification.
Identification of sleep/wake states is used in several areas of medical science. For infants, sleep state is an important variable in analyzing heart rate variability and other measures in the study of life-threatening events such as bradycardia and apnea. Polysomnography (PSG), which includes an electroencephalogram (EEG), electrooculogram (EOG) and electromyogram (EMG), is considered to be the most accurate procedure in determining sleep states. The largest shortcoming of PSG is that it is rather expensive and too complex to be used by an untrained person. Relatively high intrusiveness of the PSG method is also the cause of its low tolerance by nursing-home patients and infants. An appealing alternative is presented by the actigraphic methods. Actigraphy does not require complex equipment that has to be serviced by a trained technician and is perfectly suited to be used in home conditions. Another advantage of actigraphy is its low intrusiveness on the patient. An actigraph is a wireless portable device usually worn on a wrist or an ankle. It includes a motion sensor (an accelerometer), a microprocessor with analog/digital circuitry and a memory chip. The motion patterns are recorded throughout the day and analyzed for the information of interest. Usually, actigraphy does not aim at identifying sleep states, such as active or quiet sleep, but attempts to determine sleep–wake patterns. It provides sleep detection results comparable to those of polysomnography and behavioral response monitoring when applied to different population groups like adults, demented nursing-home patients , young children and infants, etc. Sleep/wake identifications made in adults by using actigraphy have shown 85–95% agreement rates between actigraphy
and polysomnography. This study’s aim to validate the use of actigraphy when an accelerometer normally used to determine infant’s position is also used to provide motion data for actigraphic analysis. The described method also features a different position of the accelerometer on the subjects (diaper instead of an ankle). The accuracy of the sleep/wake state prediction by a statistical method (logistic regression) is compared to that of a computational intelligence method (supervised neural network). The accelerometer was used as a part of a major Collaborative Home Monitoring Evaluation (CHIME) study which studied home infant monitors for apnea and bradycardia for over 1000 infants. Sleep/wake state information would be helpful in analyzing approximately 100 000 min of cardiorespiratory data recorded in home conditions with no available PSG scoring.
Sleep state recognition publications
Activity-based sleep–wake identification in infants (journal)
Sleep Versus Wake Classification from Heart Rate Variability Using Computational Intelligence: Consideration of Rejection in Classification Models
Sleep-Wake Identification in Infants: Heart Rate Variability Compared to Actigraphy
Reliable Determination of Sleep Versus Wake from Heart Rate Variability Using Neural Networks (conference)
Sleep State Scoring in Infants from Respiratory and Activity Measurements
Activity-based sleep-wake identification in infants (conference)