Thanks to AI, Agonal Breathing in Heart Patients can now be Detected!

Thanks to AI, Agonal Breathing in Heart Patients can now be Detected!

Healthcare Tech Outlook | Thursday, June 27, 2019

AI in HealthcareThe new AI tool developed by a group of researchers will detect agonal breathing in heart patients, greatly improving their chance of survival.

FREMONT, CA:  A groundbreaking development by a group of researchers at the University of Washington has brought a breath of relief to heart patients all over the world. According to them, the AI system can monitor heart patients and detect cardiac arrests without even coming in contact with them. During cardiac arrests, the patients experience agonal breathing, often becoming unresponsive or unable to breathe. They might even gasp for air. The AI device detects the abnormal breathing through mobile devices such as smartphones or smart speakers and calls for help.

It requires a person nearby to conduct immediate cardiopulmonary resuscitation (CPR) on the patient, a process where they are subjected to artificial ventilation combined with chest compressions to retain brain functions. However, there is no doubt regarding its role in increasing the patient’s chances of survival.

Check This Out: Top MedTech Companies

Cardiac arrests occur most commonly in the patient’s bedroom when there is likely no one to respond to the life-threatening situation. In such cases, the AI device can leverage the smart speaker systems in the house to sound the alarm or call for help. The device is designed to continuously and passively monitor the bedroom, sending alerts to people nearby. In the absence of people, the AI device can automatically notify 911 for emergency services.

The researchers leveraged past clips from bystanders who had called the dispatchers and recorded the agonal breathing to determine the need for CPR. Around 236 clips from 162 calls were used to develop a dataset of 7316 positive clips by utilizing machine learning (ML) technology. The training clips were tested at different distances to simulate situations where the distance between the patients and the devices varied. As for the negative dataset, around 7305 sound samples were derived from 83 hours of audio clips collected during sleep studies, including normal sounds such as snoring and sleep apnea. 

The teams also conducted comprehensive testing of the device during normal sleep conditions to reduce the rate of false positives and avoid false alerts. According to the sleep data, the device categorized normal breathing as agonized breathing only 0.14 percent of the time. The test was also conducted on clips collected from a sample of volunteers, which yielded a false positive rate of 0.22 percent. The researchers aim to develop apps working on the same algorithm to run passively on the home smart speaker systems.

Weekly Brief