[Pycon] [new paper] "Stefano Terna" - A Multi-Patient Data Driven Approach to Blood Glucose Prediction

[info a pycon.it]

Title: A Multi-Patient Data Driven Approach to Blood Glucose Prediction
Duration: 60 (includes Q&A)
Q&A Session: 15
Language: en
Type: Talk

Abstract: Continuous Glucose Monitoring Systems (CGMSs) allow measuring the blood glycaemic value of a diabetic patient at a high sampling rate, producing a considerable amount of data. This data can be effectively used by machine learning techniques to infer future values of the glycaemic concentration, allowing the early prevention of dangerous hyperglycaemic or hypoglycaemic states and better optimization of the diabetic treatment.  Most of the approaches in literature learn a prediction model from the past samples of the same patient, which needs extensive calibrations and limits the usability of the system. 
In this talk, I will present the results obtained by investigating prediction models trained on glucose signals of a large and heterogeneous cohort of patients and then applied to infer future glucose level values on completely new patients. To achieve this purpose, two different types of solutions will be compared, which were proved successful in many time-series prediction problems, based respectively on Non-Linear Autoregressive Neural Network (NAR) and on Long Short-Term Memory (LSTM) networks. These solutions will also be compared with three literature approaches, respectively based on Feed-Forward neural networks (FNN), autoregressive (AR) models, and Recurrent Neural Networks (RNN). While NAR obtained good prediction accuracy only for short-term predictions (i.e. with prediction horizon within 30 min), LSTM obtained extremely good performance both for short and long-term glucose level inference (60 min), overcoming all the other methods both in terms of the correlation between measured and predicted glucose signal and in terms of clinical outcome.
During the talk, it will be showcased how the models have been built with Keras and how the data preprocessing has been built with NumPy.

Tags: [u'TimeSeries', u'Keras', u'numpy', u'deep learning', u'machine-learning', u'Forecasting', u'biomedical-data-science', u'LSTM', u'signal-processing', u'time-series']