Integration of Clinical Criteria into the Training of Deep Models: Application to Glucose Prediction for Diabetic People

Standard objective functions used during the training of neural-network-based models do not consider clinical criteria, leading to models that are necessarily clinically acceptable. In this study, we look at this problem the perspective of the forecasting of future glucose values for diabetic. In this study, we propose the coherent mean squared glycemic error(gcMSE) loss function. It penalizes the model during its training not only of prediction errors, but also on the predicted variation errors which is in glucose prediction. Moreover, it makes possible to adjust the of the different areas in the error space to better focus on regions. In order to use the loss function in practice, we propose an that progressively improves the clinical acceptability of the model, that we can achieve the best tradeoff possible between accuracy and given criteria. We evaluate the approaches using two diabetes datasets, one type-1 patients and the other type-2 patients. The results show that the gcMSE loss function, instead of a standard MSE loss function, the clinical acceptability of the models. In particular, the are significant in the hypoglycemia region. We also show that this clinical acceptability comes at the cost of a decrease in the average of the model. Finally, we show that this tradeoff between accuracy and acceptability can be successfully addressed with the proposed. For given clinical criteria, the algorithm can find the optimal that maximizes the accuracy while at the same meeting the criteria.