Efficient Estimation of the Human Circadian Phase via Kalman Filtering.

Circadian rhythms play a vital role in maintaining a person's well-being but remain difficult to quantify accurately. Numerous approaches exist to measure these rhythms, but they often suffer from performance issues on the individual level. This work implements a Steady-State Kalman Filter as a method for estimating the circadian phase shifts from biometric signals. Our framework can automatically fit the filter's parameters to biometric data obtained for each individual, and we were able to consistently estimate the phase shift within 1 hour of melatonin estimates on 100% of all subjects in this study. The estimation method opens up the possibility of real-time control and assessment of the circadian system, as well as chronotherapeutic intervention.Clinical relevance- This establishes a near real-time alternative to melatonin measurements for the estimation of circadian phase shifts, with potential applications in feedback circadian control and chronotherapeutics.

Fast tuning of observer-based circadian phase estimator using biometric data.

Circadian rhythms play a vital role in maintaining an individual's well-being, and they have been shown to be the product of the master oscillator in the suprachiasmatic nuclei (SCN) located in the brain. The SCN however, is inaccessible for assessment, so existing standards for circadian phase estimation often focus on the use of indirect measurements as proxies for the circadian state. These methods often suffer from severe delays due to invasive methods of sample collection, making online estimation impossible. In this paper, we propose a linear state observer as an elegant solution for continuous phase estimation. This observer-based filter is used in isolating the frequency components of input biometric signals, which are then taken to be the circadian state. We start the design process by fixing the observer's oscillatory frequency at 24 hours, and then we tune its gains using an evolutionary optimization algorithm to extract the target components from individuals' data. The resulting filter was able to provide phase estimates with an average absolute error within 1.5 hours on all test subjects, given their minute-to-minute actigraphy data collected in ambulatory conditions.