Researchers in Boston (USA) have developed DeepGestalt, artificial intelligence capable of diagnosing around 100 rare genetic diseases based on patients’ faces, working on the basis that these disorders have characteristic traits. While the diagnoses offered by Face2Gene – the algorithm’s app – are encouraging, much work still needs to be done.
Photo credits: Nick Youngson CC BY-SA 3.0 Alpha Stock Images
Developed by US researchers, a group of algorithms by the name of DeepGestalt enables the detection of certain genetic diseases by identifying patterns in facial photos of patients, as reported by the website Allo Docteurs.
Faster and more accurate results
Doctors can sometimes spot disorders through simple facial examinations, among them Angelman syndrome, a rare disease that causes severe developmental delay and which is often characterised by pale, wrinkled skin, a large forehead, and a triangular face shape.
The process for making a reliable diagnosis can often be long and expensive, however. Artificial intelligence developed by researchers at the Boston-based start-up FDNA represents an improvement on traditional methods. Presented in a study published in Nature Medicine on 7 January, the technology can diagnose diseases faster than doctors and offers more comprehensive screening than other algorithms.
Helping AI to improve
While existing programmes also had the ability to spot genetic diseases by analysing their traits, in many cases they are designed to detect just one disorder. DeepGestalt can identify no fewer than 100, a number that will only increase, given the ability of the artificial intelligence to learn and improve its performance as its designers add patient photos and their diagnoses to the database.
DeepGestalt is the technology used in the Face2Gene app, which has yielded encouraging test results to date. When presented with facial photos, the programme provides a list of disorders that the patient may be suffering from. The artificial intelligence is also remarkably accurate in providing an initial diagnosis. When asked to diagnose Cornelia de Lange syndrome, Face2Gene correctly identified 96.88% of cases from photos, as opposed to the success rate of 75% for health professionals.
Overall, the programme is 65% accurate, though that figure climbs to an average of 91% for the first ten diagnoses on the list.
The researchers are now looking to improve their tool. Given that images in the database are mainly of Caucasian people, the system’s designers are now looking to add other types of photos to Face2Gene and enhance the app’s ability to diagnose disorders in other ethnic groups. The tool is currently based on around 150,000 photos of patients suffering from a range of genetic diseases.
Cover photo credits: Pixabay/Pexels
Contact Allianz Partners
Jan 24, 2019
Connected objects: IBM have developed a sensor that sticks onto your fingernail to monitor your health
The way that a person grabs hold of an object or carries out an everyday movement can give an insight into the state of their muscles and, more generally, their health. In order to prov [...]
Nov 8, 2018
Is this the end of tedious training courses for nursing staff? A new professional training system based on virtual reality has been tested by nurses in Rennes (France) in early October. [...]
Sep 6, 2018
US and Korean researchers have created an artificial retina that provides an effective alternative to silicon visual implants. Ultrathin, flexible and made of 2D materials, the implant [...]