Background and Aims

For an artificial pancreas (AP) to effectively limit postprandial glucose excursions, it is necessary that the insulin’s glucose lowering effect starts early in relation to meal onset. Automated and reliable early meal onset detection could therefore enhance the control outcome of APs.

A typical AP depends on continuous glucose monitoring (CGM) for insulin dosing. Because of the slow dynamics of the glucose sensing, current CGM based meal detection approaches typically exhibit a delay of 10 minutes between actual meal onset and reliable detection. In contrast, the processes of ingestion and digestion produce sounds even before meal glucose enters the blood.

Therefore, the focus of the present work is towards the early meal onset detection based on abdominal sounds (AS).

Methods

In this work we employ AS recorded in two healthy volunteers with a condenser microphone, and present an automated approach. We use the Mel-frequency cepstral coefficients and wavelet entropy as features. These features are fed to a feed forward neural network for discriminating the “meal” and “no-meal” classes

Results

This approach detects meal onset with an average delay of 4.3 minutes in our limited number of subjects. Importantly, it provides lesser delay than the state-of-the-art CGM based approach .

Conclusions

Preliminary results indicate that the AS-based approach [1] may provide early meal onset information. This can be exploited in an AP through allowable earlier meal insulin boluses, resulting in improved glycemic control.

References: