Customized Pacemakers May Learn What Our Hearts Need

Pacemakers deliver an electrical charge to reset hearts that fall out of rhythm, but different hearts could benefit from customized shocks. Rice University and Texas Heart Institute (THI) researchers are developing pacemakers tuned to the needs of individual patients that deliver more efficient signals to hearts at just the right time.
 
The researchers have received a four-year, $1.2 million National Science Foundation grant to explore the novel use of machine learning systems to precisely synchronize a pacemaker’s effect across multiple locations in the heart.
 
Earlier this year, engineering students mentored by Rice faculty and THI researcher Dr. Mehdi Razavi explored the development of a network of pacemaker nodes the size of a grain of sand that would be placed inside the heart and connected wirelessly to a central station under the patient’s skin. When the station sensed a problem with the heart’s rhythm, it would trigger the nodes to release enough energy to re-establish the heart’s normal rhythm.
 
Their work won an Excellence in Capstone Engineering Design Award at this year’s George R. Brown Engineering Design Showcase.
 
The grant allows labs led by Rice engineers Behnaam Aazhang and Yingyan Lin and their THI colleague Razavi to continue the project. They will develop hardware and software that use machine learning techniques to continuously improve a pacemaker’s function as it learns to recognize warning signs in a patient and respond with therapeutic pacing in real time.
 
In collaboration with Rice’s Joseph Cavallaro, a professor of electrical and computer engineering and computer science, and Aydin Babakhani of UCLA, a former professor and current adjunct professor at Rice, they intend to address the effects of conduction disorders that impede electrical signals in the heart and can cause fibrillation or cardiac arrest. Pacemakers are often prescribed to rebalance the signals, but Rice and THI researchers say they can be improved.
 
“Right now, if a person has a heart attack, the solution is a very big shock to the body to restart the heart,” said Aazhang, the J.S. Abercrombie Professor of Electrical and Computer Engineering. “But every patient is different, their situations change, and each episode is different.”
 
Signal processing and machine learning algorithms, a specialty of Aazhang’s lab, will learn from every cardiac episode experienced by a patient and improve their response. “Basically, the algorithm will learn about the patient’s condition on the fly and develop the ability to infer when intervention is required,” Aazhang said.
 
“A heart can be resynchronized with very low energy at multiple locations,” he said. “With pacemaker nodes in all four chambers of the heart, the energy will lead to spatially and temporally precise pacing. The amount of charge will be flexible, whatever the protocol requires.”
 
Machine learning is already playing an important role in health care to help doctors improve patient care and quality of life, according to Razavi, director of the Electrophysiology Clinical Research and Innovations department at THI.
 
“Our project could contribute significantly to overall improvement of the sensitivity of the current pacemaker algorithm,” he said. “Of course, the accuracy of any machine learning algorithm depends on the data set from which it learns.”
 
Razavi and his team will collect electrocardiogram data from a minimum of 150 patients to help train the software. Initial data will come from volunteers with wired pacemakers that doctors can monitor.
 
That will help Rice engineers update the central control unit placed under the skin that connects to multiple pacemaker nodes along wires in the heart. The unit keeps the pacemakers charged, sends and collects data and activates the nodes when triggered to pace the heart back into rhythm.
 
“We’ll make progress in stages,” Aazhang said. “First we’ll demonstrate that the controller is feasible to build. In the next phase, we’ll focus on the communication and energy-harvesting between the box (the central control unit) and the nodes, then build the nodes and so on.”
 
Lin, an assistant professor of electrical and computer engineering with expertise in energy-efficient machine learning systems, said power-hungry algorithms can affect the life of a pacemaker battery, so it will be a challenge to squeeze as much utility out of the battery as possible.
 
“In current pacemakers, the battery needs to be replaced through surgery every seven or eight years,” she said. “The next-generation pacemaker is expected to last for 15 years. One key challenge is that our proposed pacemaker must provide continuous real-time processing of data from an unprecedented number of sites simultaneously while making the hardware smaller and the battery last longer.
 
“To address this challenge, we will develop techniques to implement the algorithm with an extremely energy-efficient and ultra-miniature, application-specific integrated circuit.”
 
Aazhang noted that machine learning has entered the mainstream. “Big banks use it to analyze billions of data points to make decisions,” he said. “Hearts aren’t going to produce that much data, but now that we have the technology and the know-how, we know this is doable.
 
“I think it’s about time. Right now, our pacemakers are built with technologies from about 20 years ago. It’s time to get into the 21st century and innovate in this space.”