Chasing early signs of atriumfibrillation in fitness tracker time series data

Chasing for patterns of atrium fibrillation (AF) and heart failure (HF) in HR and HRV time series data. Sophisticated machine learning tools are applied to in debt analyse the data. Long short term memory routines take early signs into consideration and facilitate the prediction of future episodes of AF. 

This study is of importance for cardiac research because it takes behaviour into account while studying heart diseases. Predictive capabilities of the machine learning tools enables home monitoring of cardiac patients and reduces unneccessary hospital visits

Medicine safety monitoring with bio-feedback

Annually adverse drug events cause 40.000 preventable day care treatments in the Netherlands. This research project anticipates to create ML classification models to identify and predict cases of adverse drug events.

Preventable adverse drug events do charge the Dutch healthcare ecosystem an annual amount of 85 mln euro. An enormous burden for patients, care takers, insurers and at the end of the day: society. Fitness tracker data will be analysed bij machine learning algorithms and assessed by pharmaceutical experts. Potentially reducing healthcare costs and casualties.

In dept study of machine learning algorithms like 

LSTM, Autoencoder and Classification models

Sensitivity and specificity play vital roles in evaluating the accuracy and reliability of a diagnostic model. These elements co-exist and together inform the strengths and shortcomings of a model. 

The value of this study lies in the fundamental importance to find episodes in the time series data that bear the information pointing towards AF. Studies have demonstrated the possibility to detect AF from high frequency (>125Hz) PPG signals utilyzing LSTM-CNN algorithms. If AF can be detected however from simple fitness tracker data, the societal importance will be huge. Because as a cardiac disease, AF implies a substantial individual and societal burden.