The Role Of Energy-Efficient Circuits In Wearable Healthcare Applications

Challenges of designing a wearable, closed-loop seizure detection microsystem.


As beneficial as they are, health monitors for conditions like high blood pressure, arrhythmia, and epilepsy can be uncomfortable and inconvenient due to all of their protruding wires. This opens up an opportunity for designers of wearable healthcare applications.

“Wearable electronics are needed for proactive healthcare,” said Dr. Jerald Yoo, an associate professor in the Department of Electrical Engineering and Computer Science at the Masdar Institute of Science and Technology in Abu Dhabi. The ability to detect and treat chronic disease early can be a powerful countermeasure. However, this effort calls for the collection and monitoring of voluminous amounts of data, which is where wearable healthcare devices come in.

For wearable (and wireless) healthcare devices to be successful, Yoo noted, they must be energy efficient, minimally obtrusive, and disposable. Yoo discussed the role of energy-efficient circuits in wearable healthcare applications during a recent talk to Cadence employees at the company’s San Jose headquarters.

Healthcare wearables: what goes into the design?

According to the World Health Organization, about 50 million people worldwide have epilepsy. Currently, diagnosing this severe neurological disorder involves doctors interviewing the patient and administering an electroencephalogram (EEG) test, said Yoo. But these methods are hardly conclusive—what is really needed is continuous monitoring, he said.

During his talk, Yoo discussed the challenges and techniques to designing biomedical circuitry. As an example, he highlighted a closed-loop seizure detection microsystem. Creating such a system calls for several key components.

First comes the platform. Here, Yoo considers the introduction of printed fabric circuit boards about six years ago to be quite the revelation. Direct screen-printing of conductive ink on fabric has made many wearable applications possible. The technology also provides an alternative to wet electrodes (which can trigger skin sensitivities if worn for long periods) and dry electronics (which have high electrode impedance and, thus, more noise). Designers creating fabric circuit boards must address a number of challenges, including pad number limits and issues such as heat protection, static and dynamic parameter variation, and high impedance.

Next is the sensor I/F circuit—basically, these designs should use low-noise, energy-efficient implementation circuits. Yoo noted that here, it’s important to have a dedicated DC server loop to remove the electrode offset. Since the servo loop itself elevates noise, Yoo has worked with his students to create a design prototype of a wearable EEG that includes a 500Hz chopper at the servo loop for better noise efficiency.

The digital backend is also critical, providing patient-specific classification and requiring energy efficiency. In this area, there are some distinct EEG seizure detection challenges to be aware of. Namely, intra-patient age-to-age EEG variations and spatial EEG variations are unexpected outcomes. “Many times, the pattern for seizure and non-seizure is very different. The seizure pattern from patient A has almost no correlation to patient B. Other chronic diseases have similar issues,” he explained.

Machine learning addresses variations

Yoo has found that the introduction of machine learning via support vector machines (SVMs) provides a way to resolve these unexpected outcomes. There are two options here: linear SVM (LSVM), which requires limited seizure patterns but offers moderate classification accuracy, and non-linear SVM (NLSVM), which requires sufficient seizure patterns and has high classification accuracy. In his design prototype, Yoo’s choice was to use two LSVMs, one trained for sensitivity and the other trained for specificity. Using this approach in a single system, he found accuracy rates of 95% for sensitivity and 98% for specificity detection performance.

Finally, there is the system-level consideration, which, in this case, consists of the seizure detection system based on a wirelessly powered electrocardiogram (ECG). Yoo and his students used a fully integrated EEG SoC consisting of a 1.8V analog front-end and 1.0V digital backend with 16 channels, SVM and simulation, scalable EEG processing, and machine learning for patient-specific seizure detection. All of their work was implemented using Cadence tools. “System-level consideration for circuit design is very important—you don’t want to burn all of the power from the analog front end and vice versa,” Yoo noted. The next step for Yoo’s work is to test the wearable EEG design on patients.