Neural Network Identifies Possible Syndromes from Facial Profiles
A Boston biotechnology company has built a neural network which examines the overall elements of the human face and compiles a list of the ten genetic syndromes that person potentially has, possibly helping medical professionals narrow down the diagnosis for that individual. While not flawless, the algorithm is better than human efforts to emulate this technology.
The neural network, DeepGestalt, was trained on 17,000 correctly-labelled images corresponding to over 200 genetic syndromes. It was then told to identify possible genetic disorders from another 500-odd photographs of people with those conditions. The correct condition was in its top ten responses 91% of the time.
As it works, the program highlights those facial regions most useful in determining conditions, and makes them available or doctors to view. This can help medical professionals understand the relationship between genetic makeup and physical appearance.