In Australia, stroke is one among the leading causes of death and long-term disability. Some 5% deaths are attributable to stroke, while strokes cost money to the Australian healthcare system AUD 6.2 billion annually.
Strokes occur when there’s a sudden lack of blood flow within the brain. This prevents brain tissue from getting the oxygen and nutrients it needs, which can lead to damage to parts of the brain.
Timely stroke treatment can reduce brain damage and improve patient outcomes. However, this relies on early recognition of symptoms, which just isn’t at all times easy.
Our team has developed a brand new smartphone app that can check an individual’s facial features and detect whether or not they are having a stroke. We recently published the outcomes of study A pilot study this tool and found that it could quickly and relatively accurately discover whether someone had had a stroke.
Facial expression scanning
One of the earliest outward symptoms of a stroke is a facial features, comparable to drooping, when one side of the mouth just isn’t activated when the person tries to smile.
However, paramedics responding to emergencies and hospital emergency ward stroke cases are sometimes neglected by staff. Facial expressions naturally vary from person to person, and recognizing subtle changes in high-stress environments is difficult. This could also be even tougher if the patient is from a unique ethnic or cultural background.
With our smartphone app, a paramedic or other first aider asks the patient to try to smile and “films” the patient’s face while she or he does so. The AI-powered model then analyzes the video footage to search for symptoms similar to those who doctors use to discover a stroke, namely asymmetrical mouth drooping.
The application has been designed with simplicity in mind – just point the camera on the patient and press the button. To ensure patient privacy, the video is analyzed in real time and doesn’t require storage. This device would only require a smartphone, so it could be easy to implement and can be a cheap solution.
The idea is that first responders, comparable to paramedics or emergency department nurses, can have this app on their smartphones. When they first see a patient who has experienced a medical emergency, they can use the app inside seconds to detect whether the patient could have suffered a stroke. In this manner, treatment can be accelerated accordingly.
Our pilot study
We tested the tool on a small dataset, using video recordings of 14 individuals who had experienced a stroke and 11 healthy controls.
We found that it was 82% accurate, meaning it appropriately identified stroke 82% of the time. Our tool doesn’t replace comprehensive clinical stroke diagnostic testing, however it can help discover individuals who need treatment far more quickly and assist clinicians.
Although these results are promising, we plan to further optimize the model. We hope that accuracy will improve as a bigger dataset is created that features recordings from more patients.
At this stage, the AI model has only been trained and developed on a small dataset, and the info lacks diversity by way of ethnicity and demographics. It can be obligatory to refine and test the applying for people from different cultural and ethnic backgrounds.
In the longer term, we plan to cooperate with doctors, emergency departments and ambulance services to conduct clinical trials. We will need to test the effectiveness of this tool within the hands of real users, comparable to paramedics, to confirm that it helps them look after patients.