25 January 2021

Two years after the start of SINTEC project, the activities related to Comparative Electrophysiological Readouts keep going on. The activity carried out so far in this field of research focuses mainly on physiological parameters taken from subjects under examination. Specifically, on the estimation of Systolic blood pressure (SBP) and Diastolic blood pressure (DBP) starting from electrocardiographic and photoplethysmographic signals.

The approach that has returned a better result in relation to computational load, it is the approach based on regressive analysis.

While waiting to receive SINTEC modules, since the goal is to obtain results in real time, signals from the MIMIC III online database were used and their acquisition in real time was simulated.

It is an approach that requires an initial calibration in which signals will be taken from subjects under examination for an estimated time of a few minutes. This happens through a precise protocol. After this period the algorithm will be person-specific and will therefore be able to estimate the blood pressures of the subject.

Results obtained with this approach are good but can be improved. Particular attention is given to the Mean Absolute Error (MAE) value which is the value used by current ANSI/AAMI/ISO requirements. It is tolerable a maximum MAE of 10 mmHg for at least the 80% of occurrences.

As mentioned above, this approach can be improved. M24 work is based on this. Among the various steps taken to improve the performance of system, in a first place the regression equation was modified, which however did not lead to a great improvement; secondly, corrective factors were used. These corrective factors aim to tighten the error distribution bell.

It was also evaluated how to avoid overfitting by taking into account a smaller number of cardiac cycles that are however sufficiently informative. In this limited number of cycles, for example, situations must arise in which subject is not only in the rest phase but also in the light activity phase.

Approaches are also being sought that allow for better results without increasing initial calibration times. Specifically, we are trying to have better quality training by evaluating the standard deviation of results. For the same purpose, it is ensured that the best quality data carries a greater weight in training.

In the image above it is possible to notice some differences obtained with the use of corrective factors that go to decrease the error on prediction of the pressures.

In the image above it can be seen that the error for systolic pressure has greatly decreased after the application of corrective factors. SBP is in fact usually has a higher MAE value due to the high variability of its values.