Digital camera and AI algorithm can detect facial palsy

Researchers from Middle Technical University (MTU) and the University of South Australia (UniSA) have developed a new diagnostic tool using artificial intelligence and a digital camera to detect facial palsy with high levels of accuracy. The researchers say their tool, described in the journal BioMedInformatics, can reduce diagnostic errors that often occur with this common and treatable neurological disorder.

Facial palsy is caused by impairment of the facial nerve, resulting in temporary muscle weakness or paralysis on one side of the face, affecting approximately one in 60 people worldwide during their lifetime. The researchers say that detection based on visual examination can be inaccurate because facial palsy often mimics other conditions and can also be subtle in its presentation, with a 2020 study estimating that misdiagnosis occurs in up to 20% of cases. Detection followed by investigation is important because possible causes include stroke, HIV infection, multiple sclerosis, Guillain-Barré syndrome and Lyme disease.

The researchers developed a real-time detection system for facial palsy using a Raspberry Pi microcomputer, a digital camera and a deep learning algorithm. Using a dataset of 26,000 images, containing 19,000 normal images and 1600 facial palsy images, the researchers employed AI techniques to train computer vision systems to recognise the condition, differentiating them from healthy individuals. They then took photos of 20 patients with different degrees of facial palsy, using an algorithm to detect the condition in real time, as well as identifying their approximate age and gender.

UniSA remote sensing engineer Professor Javaan Chahl said the system achieved a 98% accuracy rate. He noted, “Using computer vision systems to detect facial palsy could not only prevent misdiagnosis, but also save patients and medical specialists time, effort and cost.”

The breakthrough comes just weeks after MTU and UniSA researchers revealed their promising results in computer-aided disease diagnosis based on tongue colour, which they used to successfully identify disease (either diabetes, renal failure or anaemia) in 94% of patients.

Image credit: iStock.com/Doucefleur