Wrist Actigraphy Algorithms: The Key to Understand Sleep

Wrist actigraphy has become a popular non-invasive and cost-effective method for monitoring sleep and physical activity patterns. With the increasing use of smartwatches, wrist actigraphy has gained even more popularity. Several algorithms have been developed to accurately analyze wrist activity data and classify sleep and wake periods. In this post, we will explore the top algorithms used for wrist actigraphy analysis.

Cole-Kripke Algorithm

The Cole-Kripke algorithm was developed by Cole and Kripke in 1991 and has since become a widely used algorithm in sleep research.
The algorithm works by first filtering the wrist activity data to remove any high-frequency noise. The filtered data is then analyzed to determine the activity counts for each minute. The activity counts are then converted into a binary form by setting a threshold value above which a minute is considered active and below which it is considered inactive. The threshold value is calculated based on the mean and standard deviation of the activity counts for each individual. The mean activity count is subtracted from each minute’s activity count, and the resulting value is divided by the standard deviation. The threshold value is set to a certain number of standard deviations above the mean, which can vary depending on the study’s requirements. Once the data has been converted into binary form, the algorithm applies a set of rules to identify periods of sleep and wake. The rules are based on the number of consecutive minutes of activity and inactivity and the time of day.
For example, if a period of inactivity lasts for a certain number of consecutive minutes during a particular time of day, it is classified as sleep. The algorithm also incorporates a set of heuristic rules to refine its classification of sleep and wake periods. These rules take into account factors such as the length of the sleep period, the proportion of sleep and wake periods throughout the night, and the amount of time spent in bed.

Sadeh Algorithm

The Sadeh algorithm was developed by Avi Sadeh in 1994. This algorithm’s working principle is almost the same as that of the Cole-Kripke algorithm. The Sedah algorithm also has a second version named Sedah 2.

Tudor-Locke Algorithm

The Tudor-Locke algorithm was developed by Catrine Tudor-Locke in 2002. The algorithm works by first filtering the wrist activity data to remove any high-frequency noise. The filtered data is then analyzed to determine the activity counts for each minute. The activity counts are then classified into one of four categories:

Sedentary: This category represents activities that involve little or no movement, such as sitting or lying down.
Light: This category represents activities that involve some movement, such as slow walking or standing.
Moderate: This category represents activities that involve moderate movement, such as brisk walking or cycling.
Vigorous: This category represents activities that involve high-intensity movement, such as running or jumping.

The classification of activity counts into categories is based on threshold values that are set for each category. These threshold values are typically based on calibration studies that use both wrist activity data and energy expenditure measurements to determine the threshold values that correspond to each activity category. Once the activity counts have been classified into categories, the algorithm calculates various physical activity metrics, such as the time spent in each activity category, the number of bouts of physical activity, and the intensity-weighted activity counts. These metrics can be used to estimate an individual’s overall physical activity level and to assess changes in physical activity over time.

Actiware Algorithm

The Actiware algorithm was developed by Philippe Jean-Louis and their colleagues in 1999 and is commonly used in clinical settings. This algorithm’s working principle is almost the same as that of the Cole-Kripke algorithm and the Sedah algorithm. The Actiware algorithm also includes a feature called sleep stage scoring, which uses additional sensors to estimate the sleep stage (i.e., deep sleep, light sleep, and REM sleep) during sleep periods.

Bayes Actigraphy Sleep Algorithm (BASA)

The Bayes Actigraphy Sleep Algorithm (BASA) is a machine learning-based algorithm that uses Bayesian modeling to accurately classify sleep and wake periods based on wrist actigraphy data. It was developed by researchers at the University of Michigan and is widely used in both clinical and research settings.
The BASA algorithm works by first processing the raw wrist activity data to extract a set of features that are indicative of sleep and wake periods. These features include measures such as the duration and frequency of wrist movements, the rate of change of wrist activity, and the regularity of wrist activity patterns. Once these features have been extracted, the BASA algorithm applies a Bayesian modeling approach to classify each minute of wrist activity as either sleep or wake. Bayesian modeling is a statistical technique that allows for the incorporation of prior knowledge and assumptions into the classification process, which can improve the accuracy of the algorithm.
The algorithm uses a personalized approach to sleep classification that takes into account the individual’s typical sleep patterns and adjusts the classification criteria accordingly. This personalized approach helps to improve the accuracy of the algorithm by accounting for individual variability in sleep patterns. The algorithm includes a set of rules that help to eliminate false positive and false negative classifications, such as by excluding periods of high activity that may occur during wake periods or by adjusting the classification criteria based on the length of the sleep period.

Adaptive Sleep-Wake Classifier Algorithm (ASWC)

The Adaptive Sleep-Wake Classifier (ASWC) Algorithm was developed by researchers at the University of Surrey. It’s basically similar to BASA with a different modeling approach.

Conclusion

While wrist actigraphy algorithms have already contributed significantly to our understanding of sleep and physical activity, ongoing research and development in this field will be crucial for improving accuracy and reliability. So, the question remains: what new insights will we uncover about sleep and physical activity through wrist actigraphy?