Fate of viral content on the intent can be dichotomous: they either win the hearts of many or become conspiracy. The latest victim being “#10yearschallenge”. The hashtag took over the major social media channels at the turn of the year. The result was not only a decade old profile picture placed alongside with those from year 2019. There was a speculation that Facebook might have been secretly collecting data to train its facial recognition tool.
The hypothesis creator later claimed that she was being sarcastic. But our associative memory has taken us to believe our data and privacy are being exploited unknowingly. Coincidentally, around the same time, Nature published an article detailing the deep learning algorithm behind Face2Gene.
Face2Gene is a mobile application used mainly by doctors to underlie patients with genetic disorder via their physical appearances. The mobile application has been in use for the past few years and it is now being systematically tested to see if it truly helps doctors in their practice.
A form of diagnostic aid
Rare genetic disorders may change one’s physical appearances, as most of them hinder normal growth and development. From upward-slanting eyes common in Down Syndrome, to short nose and downturned corners of mouth present in Cornelia de Lange syndrome. Individual differences sometimes make it challenging for doctors to provide accurate diagnoses.
That’s when doctors will encourage patients to submit a photo of themselves and verify using Face2Gene to confirm the diagnosis. According to the published results, 91% of the time, the deep learning algorithm DeepGestalt, is able to place the right diagnosis on its top ten list. It has outperformed doctors in diagnosing disorders such as Cornelia de Lange syndrome and Angelman syndrome.
On patient’s end, such diagnostic aid also saves them from additional medical bills spent on getting multiple genetic tests. Apart from genetic disorders, researchers had been trying to employ facial recognition to assess basic physiological health or even wound care.
Challenges and developments ahead
Like most facial recognition tools of similar nature, insufficient training data and racial bias are in the way. In the case of Face2Gene, the small population of patients affected by genetic disorders is putting a threat to its potential. Likewise, training algorithm with mostly Caucasian faces may make it potent to diagnose patients from different ethic backgrounds.
The other challenge is whether the technology can neatly assimilate into the practice. For example, dermatologists may laugh when data is asked. Because in dermatology, even if it is remotely treated, there is usually no image or related records taken. The physicians will assess everything visually and write them down in words. Hence, if researchers will ever like to obtain dermatology data to train fellow artificial intelligence (AI) to recognize various skin conditions, physicians may first need to change the way they provide care. Even if this has somehow been overcome, still dermatologists may not adopt AI since it never exists in their workflow.
As such, whilst conspiracy is not likely, winning the hearts of clinicians and patients are definitely part of the developments concerning facial recognition in medicine.
A science writer with data background and an interest in current affair, culture and arts; a no-med from an (almost) all-med family. Follow on Twitter.