Getting your Trinity Audio player ready...
|
In collaboration with other researchers, scientists from the National Institute of Standards and Technology (NIST) have utilised machine learning, a type of artificial intelligence (AI), to detect abnormal cardiac rhythms among firefighters.
Their study aims to develop a portable heart-monitoring device that firefighters can wear. This device would help identify early indicators of heart problems, allowing firefighters to seek medical assistance on time, potentially preventing severe consequences.
In 2022, the National Fire Protection Association reported that 36 firefighters died on duty due to sudden cardiac death. This tragic occurrence transpires when an irregular heart rhythm disrupts the heart’s ability to pump blood, often linked to a heart attack.
The incidence of sudden cardiac events among on-duty firefighters is twice as high as that of police officers and four times higher than other emergency responders.
NIST researcher Chris Brown emphasised the alarming trend, “Year after year, sudden cardiac events remain the leading cause of death among firefighters”. These events result in fatalities and lead to career-ending injuries and long-term disabilities for affected firefighters.
Firefighters operate in extremely demanding environments, engaging in physically strenuous tasks such as carrying heavy loads, navigating staircases, and enduring extreme temperatures. These challenging conditions limit their cooling ability, leading to significant discomfort. Unfortunately, reports indicate that firefighters often attempt to persevere through these circumstances without recognising the potential risk of sudden cardiac death.
In response to this concern, researchers from NIST collaborated with experts from the University of Rochester School of Nursing. Around ten years ago, Mary Carey, a researcher from Rochester, and her team collected 24-hour electrocardiogram (ECG) data from 112 firefighters.
These firefighters wore chest-mounted electrodes, enabling the researchers to record their heart activity during 16-hour on-duty and eight-hour off-duty periods. The data encompassed various activities undertaken by firefighters, including responding to fire and medical calls, exercising, eating, resting, and sleeping.
Rochester co-author Dillon Dzikowicz expressed the uniqueness and significance of the collected firefighter data, acknowledging its crucial role in advancing their research efforts and safeguarding firefighters.
Leveraging the extensive dataset from Rochester, the researchers employed machine learning techniques to develop the Heart Health Monitoring (H2M) model. They utilised segments of 12-second durations from a substantial portion of the ECG data to train the H2M model. Within the ECGs, individual heartbeats were categorised as normal or abnormal, indicating irregular heart rhythms such as atrial fibrillation or ventricular tachycardia.
NIST guest researcher Jiajia Li explained that the model was designed to effectively identify and learn ECG patterns from normal and abnormal heartbeats, enabling accurate detection of irregular cardiac rhythms.
After the training and validating of the H2M model, it was employed to analyse firefighter ECG data from the previously unseen portion of the Rochester dataset. When confronted with around 6,000 abnormal ECG samples, H2M demonstrated a remarkable accuracy of approximately 97% in correctly identifying these abnormalities.
To ensure reliability, H2M was also trained using ECG datasets from individuals who were not firefighters. However, when applied to the firefighter data, H2M exhibited an error rate of roughly 40% in detecting cardiac events.
NIST researcher Wai Cheong Tam highlighted the significance of utilising the appropriate dataset for training the AI model, emphasising its critical role in achieving accurate results.
The researchers envision integrating this model into portable heart monitors that firefighters can wear during their duties. This real-time AI assistant would provide timely alerts regarding cardiac irregularities, serving as a valuable resource akin to having a cardiologist accompanying the firefighting team.
Tam expressed the potential life-saving impact of this technology, underscoring its ability to benefit firefighters, other first responders, and various segments of the public. Training the AI with relevant ECG datasets could expand this approach to assist other groups and populations.