Deliverable 3.2 - Unobtrusive solution for the monitoring and the assessment of physical status in adults with CHF

This deliverable begins with a description of the monitoring devices used in the HeartMan system. The main device is the HeartMan wristband, which measures the heart rate, heart rate variability, photoplethysmogram, galvanic skin response, temperature and acceleration. It can connect to the mobile phone via Bluetooth low energy. Additional devices are the Ruuvi ambient sensor, which monitors the temperature, humidity and light at home, and also connects to the mobile phone via Bluetooth low energy; a blood pressure monitor, scales and pill dispenser. The acceleration and heart rate from the wristband are used to recognise the patients' activity and estimate their energy expenditure. To do this, the sensor data are split into windows, a number of features (such as averages, standard deviations and angles) are computed from each window, and finally these features are fed into a model built with a machine-learning algorithm to recognize the activity or estimate the energy expenditure. The HeartMan method can recognize ten activities with the accuracy of 72 %. The error of the energy expenditure estimation is 0.58 metabolic equivalents of task, which is better than dedicated consumer devices. The photoplethysmogram is used to estimate the blood pressure. This is a challenging task that requires complex processing. The data are first cleaned, and cycles (corresponding to heartbeats) are extracted. A number of features are extracted from each cycle, which - together with the raw signal - are fed into a deep neural network to estimate the blood pressure. Some data belonging to each patient is required for personalisation, to achieve adequate accuracy. The error on hospital data is below 8 mmHg for systolic blood pressure, and below 4 mmHg for diastolic blood pressure. The errors on real-life data, where traditional regression was used instead of deep learning because of lack of data, are around 12 and 6 mmHg.