Regularized Denoising Method for Retrospective Detection of Pressure Induced Artifacts in Continuous Glucose Monitoring Sensors Data.

Background: Retrospective analysis of Continuous Glucose Monitoring (CGM) sensor data can play an important role in improving glucose control and driving therapy adjustment in clinical settings. To prevent incorrect clinical decisions, however, preliminary detection and elimination of CGM data portions affected by errors and artifacts is of paramount relevance.

Objective: This paper deals with the retrospective model-based detection of Pressure Induced Sensor Attenuations (PISAs) in CGM data, which could be misinterpreted as hypoglycemic events.

Methods: In a Bayesian framework, we proposed a method that, to detect PISAs, leverages CGM data and a-priori statistical information on the expected smoothness of the CGM signal and the measurement error affecting it. The proposed strategy's effectiveness is evaluated using an in-silico dataset, generated by the FDA-accepted UVa/Padova Type 1 Diabetes simulator, and a real-world dataset, gathered using a commercial (Dexcom G6) sensor.

Results: For the simulated data, the PISAs detection performance achieves a sensitivity of 61.5%, with 0.24 false positives per day. In the real-world dataset, the method exhibits a sensitivity of 57.3%, with 1.15 false positives per day.

Conclusions: These results demonstrate the potential of the proposed approach for retrospective detection of PISAs in CGM data.

Significance: By removing artifacts, CGM data quality can be improved before its retrospective use for diabetes therapy adjustments.