Wearing Your Heart (Monitor) on Your Sleeve

The diagnosis and treatment of complex medical disorders begins with data. Traditionally this data has been captured through examinations, laboratory tests, or imaging procedures. Unfortunately, these types of tests most often require that a patient visit a clinic where data is collected under unnatural conditions. Moreover, these clinic visits only allow for the intermittent capture of data. This presents physicians with many challenges as they attempt to fully characterize the condition of an individual using a source of information that is very incomplete.

Wearable electronic devices have the potential to overcome these issues by continuously monitoring physiological parameters to provide medical professionals with a complete picture of their patients’ health. Consider the monitoring of blood pressure, for example. This would normally be done during an office visit by using a cuff. Blood pressure is highly dynamic, however, so a single measurement taken while sitting in a chair is of limited value.

Recognizing that hypertension is a big factor in cardiovascular disease — the leading cause of death worldwide — a team of researchers at Nanjing University set out to explore the state of the art in wearable cuffless blood pressure monitoring. In particular, they looked into the latest advances in flexible electronics and machine learning to determine what advances may be useful in building a practical monitor, and also what areas still need further work to get us over the goal line.

In terms of sensing options, the researchers found that flexible mechanical sensors are a common choice. These low-cost sensors can be accurate, but require very precise positioning on the body, which can impact that accuracy. Some mechanical options, like triboelectric sensors, are self-powered, which is highly desirable in a wearable. Optical sensors, especially those that use photoplethysmography, are also a good candidate. These sensors are highly accurate, but tend to be larger, and therefore less comfortable to wear. Flexible ultrasound sensors are also very accurate, but much like optical sensors, they tend to be on the large side.

Translating the data captured by these sensors into an accurate blood pressure measurement is challenging, but a number of machine learning techniques are simplifying the process. One promising technique involves the use of convolutional neural networks (CNNs), which excel in feature extraction by applying different filters to the data during training. CNNs consider the spatial structure of data, making them effective in identifying local features important for blood pressure estimation.

Recurrent neural networks (RNNs) are particularly useful for processing sequential data, as they take into account the temporal order of data and maintain connections between different points in the sequence. This makes RNNs effective in preserving past information, which is crucial for continuous blood pressure monitoring. However, training RNNs can be more complex, particularly when dealing with long sequences, where issues like vanishing or exploding gradients can arise.

At the conclusion of their review, the team found that there are still some areas that need to be addressed before wearables are truly ready for clinical applications. The flexible sensor measurements can be negatively impacted by factors like the user’s posture, or changes in environmental factors like temperature and humidity, for example. Furthermore, the machine learning models used to estimate blood pressure lack universality and may not work for all individuals.

It is the hope of the researchers that by shining a light on both the strengths and weaknesses of existing technologies, advances will be made that enable the development of practical, wearable cuffless blood pressure monitors in the near future.