Abstract:
|
We visualize patients' blood glucose patterns using a two-step approach. In step one, we use B-spline based model to extract features from continuous glucose monitoring (CGM) data. This step allows us to transform varying length data to fixed length features with more than 100 free parameters. In step two, we use t-Distributed Stochastic Neighbor Embedding(t-SNE) method to reduce the features to two or three dimensions data. The resulting visualization gives more detailed clustering than naive methods such as plotting mean versus standard deviation. We illustrate this approach using CGM data from more than 100 Type 1 diabetes patients collected in the A1C-Derived Average Glucose (ADAG) study.
|