Cancellation of Movement Artifacts in GlucoseSensor Data

Abstract

The BioMKR wearable device developed by Prediktor Medical is aiming to be the first non-invasive continuous glucose monitoring device on the consumer market. The data gathered by the device, however, contains movement artifacts that limit its performance. The artifacts responsible for this deviation is believed to be correlated with the external forces acting on the device. Therefore, the use of a force sensor as an auxiliary measurement is evaluated. A Force Sensing Resistor is integrated with the BioMKR circuitry to enable stand-alone logging of data. The main purpose for this study is to evaluate the possibility of using the force measurements to characterize the signal perturbations, and use the data set to compensate for faulty readings.

This thesis evaluates the influence the external forces captured by the FSR sensor has on the recorded near infrared signals. Two prototypes have been tested, where the second prototype needed additional improvement of the FSR actuating system. Different adapters for the actuating system was developed, and their performance are measured with respect to FSR signal stability, sensitivity and skin irritation. Test conducted with the prototypes replicate real use of the device, and consists of strap test, night test, temperature test and external force test.

The results from night testing have been modeled using PLSR, which showed some predictive properties. Building a model from one session and using it to predict the FSR from another session proved unsuccessful, implying weak correlation between the FSR signal and NIR signal. The alteration of the optical properties of the tissue as a reponse to applied pressure is believed to affect the NIR signals in a way that the FSR alone is not able to model.

Using the force signal as an event marker for NIR changes, however, proved to be very good. It can be shown that the FSR successfully captures every significant NIR change as a result from involuntary movement during sleep.

It is advised that the force signal should be used together with accelerometer and other BioMKR parameters in determine the level of trust the resulting NIR signal has in estimating the correct blood glucose. Further evaluation on filtering methods to compensate for erroneous BioMKR data is also advised.