AUTOMATIC IDENTIFICATION OF TRAINING PHASES FROM HEART RATE RECORDED BY WEARABLE SENSORS

All the athletes in the world, in particular professional athletes, strive for success in practice and competition; therefore, the physiological monitoring of the training is becoming increasingly crucial because it can be used to better understand the underlying processes of performance and to subsequently increase or optimize it. Moreover, nowadays, the incidence of sudden cardiac death (SCD) in athletes ranges from 1:40.000 to 1:250.000 depending, among other things, on the sporting discipline. More than 80% of SCDs occurs during intensive training and competition. Therefore the need to have a tool able to help athletes to optimize their performances and, at the same time, to monitor their health status, is very important. Hence the idea to develop an algorithm able to perform the above-mentioned actions. In order to combat the SCD, the avanced algorithms for the monitoring of the athlete must be able to correctly and reliably identify training phases, particularly when it has to do with wearable sensors. The most important and used wearable sensors monitoring the cardiovascular signals are listed knowing their strengths and weaknesses. Nowadays, on the market there are textile electronics, chest straps, patches, smartwatches, wristbands and some other products such as ear-rings, ear-phones or rings. Data used for the development of the algorithm have been collected by a chest strap called Zephyr Bioharness 3.0. Some automatic algorithms for the identification of training phases are then analyzed, such as one based on the heart-rate variability, Dynascope and HeartScope. The key principle of the new developed algorithm is the analysis of the angular coefficient based on both the heart-rate and tachogram signals. On the trend of this new index, all the time-instants, needed for the identification of the training phases, are determined. The study population comprises ten middle-running athletes, all in good state of health with ages ranging from 18 and 55 years old. As regards the validation and the reliability of the algorithm, it shows a percentage of error, for both the identification of the two transition phases (from pseudo-resting to exercise, and from exercise to recovery), ranging from 0% to 1.57% in 6 subjects over 9. The remaining three subjects are characterized by a wrong identification of the exercise-recovery transition phase, despite their resting-exercise transition phase are correctly identified. The identification of the these two transition phases, if deleted from the signal for a further processing, allows the analysis of some important heart risk indeces such as QT-interval or heart-rate variability. In conclusion, the algorithm developed in this work turns out to be a reliable tool for both the optimization of athlete’s performance and the monitoring of the health status of the athlete. Moreover, although the algorithm is specifically developed for the the analysis of data coming from middle-running athletes, it has a basic structure promising to adapt to any sport. In short, a tool already reliable, but with still extensive room for improvement. A tool extremely attentive to the needs of any athlete.