Using Signal Processing Techniques to Predict Postprandial Glucose
Background & Aim:
The author has collected a complete set of post prandial glucose (PPG) and lifestyle data for a period of 994 days with 2,982 meals (6/11/2015 – 3/1/2018). This paper discusses the methodology and accuracy of his developed PPG prediction model using signal processing techniques.
Materials & Method:
Due to his academic background in mathematics, physics, and engineering, he views these biomedical and lifestyle data as a collection of nonlinear signal waves. He applied signal processing to decompose this time-series measured PPG signal into multiple (> 10 lifestyle factors) single-sourced composite waveforms, examined each composite signal, and then recombined them into a predicted PPG curve. Finally, he compared this predicted signal against the measured signal to calculate its accuracy and correlation. He further improved his model via a trial-and-error “curve-fitting” method.
The PPG’s major creation source, corresponding glucose, and contribution level are as follows:
Carbs/Sugar: 14.5 mg/dL, 37%
Post-meal Exercise: -15.7 mg/dL, 41%
Weather: 3.8 mg/dL, 10%
Measurement delay: -2.4 mg/dL, 7%
Others: -1.9 mg/dL, 5%
During this period, his averaged PPG values are:
Predicted: 119.16 mg/dL
Measured: 119.88 mg/dL
with 99.4% linear accuracy and a high correlation of 69%.
The quantitative results from the developed PPG prediction model reflect the accuracy and applicability for type 2 diabetes control via a guided lifestyle management. The utilization of signal processing from electronics engineering and computer science is also proven quite effective for this investigation.