top of page

Diabits Accuracy 

The Diabits algorithm performed with 93.6% accuracy for OhioT1DM data set.

We use Deep Learning methods to analyze  and  model glucose metabolism. Our approach takes advantage of Neural Networks ability to "learn" and "think" like a pancreas.

Diabit_Image_-14.png
Anchor 2

Dataset

iPhone Mockup 01 - Dark graph.png
Diabits_Web_drop-01.png

The OhioT1DM was developed for blood glucose level prediction research. The dataset consists of 8 weeks of continuous glucose monitoring via Medtronic sensors and self-reported life-event data for 12 people with type 1 diabetes.

Diabits%2520Logo%2520Vector_Artboard%252
Screen Shot 2021-11-13 at 6.18.03 PM.png

This dataset is chosen because in addition to blood glucose values, carbohydrate and insulin events are also recorded. 

Anchor 3
Anchor 4

Results

We recorded the sensitivity (true positive rate), specificity (true negative rate), and accuracy of our predictions for previously unseen patients. 

The model predictions and the actual blood glucose values were given a label, post prediction. The labels are based on ADA’s recognized blood glucose ranges.

Screen Shot 2021-11-13 at 9.37.55 PM.png

93.6% Accuracy

96.1%  Specificity

83.6% Sensitivity 

The labels for actual and predicted values were used to calculate the accuracy, specificity, and sensitivity. 

Anchor 5

Predicted vs. Actual 

Below is the partial overlay of actual vs predicted blood glucose values. 

Screen Shot 2021-11-14 at 5.17.07 PM.png
bottom of page