This 2013 photo was taken by nurse scientist Becky Vincent and Mary Carrie of the University of Rochester collects ECG data from firefighters at the Dewey Street Fire Department in Rochester, New York. Now, ten years later, researchers at NIST, Rochester, and Google have used this data to train an AI model to predict cardiac events.
credit:
Karen O’Hearn/University of Rochester School of Nursing
Firefighters routinely risk their lives in dangerous situations, but most deaths on the job are not directly caused by fire or smoke inhalation. Rather, about 40% of work-related deaths are due to sudden cardiac death.
Now, researchers at the National Institute of Standards and Technology (NIST) and their colleagues have used a form of AI known as machine learning to pinpoint abnormal heart rhythms in firefighters. The researchers say their work could ultimately lead to the development of a portable heart monitor that firefighters can wear to spot early warning signs of heart disease and seek medical attention before it’s too late. I hope that.
The team, which includes researchers from NIST, the University of Rochester, and Google, has fire journal.
According to the National Fire Protection Association, sudden cardiac death will kill 36 firefighters on duty in 2022. Sudden cardiac death occurs when the heart stops pumping blood due to an irregular heart rhythm, the most common cause being a heart attack. A firefighter’s on-duty mortality rate from sudden cardiac events is twice as high as that of police officers and four times as high as that of other paramedics.
“Sudden heart attacks are by far the leading cause of firefighter deaths each year,” said NIST researcher Chris Brown. “Heart disease can also cause career-ending injuries and long-term disabilities.”
A group of career firefighters from First Rescue Company in Buffalo, New York, receive a fire call. These firefighters wore electrodes and recorded electrocardiogram data (ECG) over a 24-hour period.
credit:
Mary Carey, University of Rochester School of Nursing
Firefighters work in extremely harsh environments, carrying heavy objects, climbing stairs, and withstanding extreme temperatures with limited cooling capacity. And while firefighters may experience significant discomfort, they often try to navigate through these situations without realizing they may be at risk of sudden cardiac death. It is shown.
To address this issue, NIST researchers contacted colleagues at the University of Rochester School of Nursing. A decade ago, Rochester researcher Mary Carey and her colleagues collected 24-hour electrocardiogram (ECG) data from each of 112 firefighters who wore chest electrodes. ECG data included 16 hours of work duty and 8 hours of non-work duty during which firefighters engaged in routine activities such as responding to fire and medical calls, exercising, eating, resting, and sleeping.
“The firefighter data we collect is very unique,” said Rochester co-author Dillon Zikowitz. “Having robust data is essential to advancing our work and protecting firefighters.”
The researchers then used machine learning and the Rochester dataset to build what they called a heart health monitoring (H2M) model. They trained his H2M using most of his 12-second segments of ECG data. Individual heartbeats in the ECG were classified as normal heartbeats or abnormal heartbeats indicating irregular heart rhythms such as atrial fibrillation and ventricular tachycardia.
“The model is designed to effectively learn ECG patterns from both normal and abnormal beats,” said NIST Visiting Scientist Jiajia Li.
A nursing scientist applies 12-lead electrodes during a physical examination to measure the electrocardiogram of a firefighter.
credit:
Mary Carey, University of Rochester School of Nursing
Once H2M was trained and validated, we analyzed the never-before-seen ECG data of firefighters from the Rochester dataset. When presented with about 6,000 abnormal ECG samples of her, H2M accurately identified them with about 97% accuracy. As a check, H2M was also trained using her non-firefighter ECG dataset. Using this non-firefighter data, H2M showed an error rate of about 40% in identifying cardiac events in the firefighter data.
“It was important to use the right dataset to train the AI model,” said NIST researcher Wai Cheong Tam.
In the future, the researchers envision incorporating the model into a portable heart monitor that firefighters can wear while on duty to provide real-time warnings of heart problems. AI assistants like this could be the next best thing for cardiologists accompanying firefighters.
“This technology can save lives,” Tam said, adding that the approach could be extended to help other groups if the AI were trained on the right ECG dataset. added. “It has the potential to benefit not only firefighters, but other first responders and even more of the general public.”
Papers: Jiajia Li, Christopher Brown, Dillon J. Dzikowicz, Mary G. Carey, Wai Cheong Tam, Michael Xuelin Huang. Toward real-time cardiac health monitoring in firefighting using convolutional neural networks. fire journal. Published online on June 28, 2023. DOI: 10.1016/j.firesaf.2023.103852
