In near future, the standard medical checkup will include a facial recognition scan. Researchers have found new algorithms to detect facial characteristics associated with genetic disorders. These algorithms will speed up the process of clinical diagnoses.
FDNA published details on their new software called “DeepGestalt” in “Nature Medicine” which is a medical journal. They have analyzed a dataset of faces to train their algorithms. They used an app called “Face2Gene” to collect over 17,000 facial images. These images cover 200 different syndromes.
They used “DeepGestalt” for searching particular disorders called “Angelman syndrome” and “Cornelia de Lange syndrome”. These disorders directly affect a person’s mobility and intellectual development. A person with Cornelia de Lange syndrome has some clearly noticeable facial traits such as curved eyebrows that join in the middle while a person with Angelman syndrome has abnormal fair hair and skin.
When identifying the patients with syndromes using their pictures, DeepGestalt recorded over 90% accuracy rate. This rate beat the professional physicians’ rate which was about 70% on the same tests. 502 pictures showing persons with 92 various syndromes were used for this testing. First DeepGestalt guessed 10 possible diagnoses for a picture and then recognized the target syndrome from these guesses over 90% of the time.
DeepGestalt algorithm was asked to recognize the genetic mutations caused for Noonan syndrome by showing it pictures of persons with that syndrome. In this experiment, the algorithm showed only 64 % accuracy rate. But if the human guesses the mutations the accuracy rate will be 20%. Therefore, the software still performed better than the human.
However, professionals in this field are not satisfied with these results. They tell that these kind of software tests are not a simple solution for diagnosing uncommon genetic disorders. Dr. Bruce Gelb who is a professor at the Icahn School of Medicine at Mount Sinai, as well as a highly skilled professional in Noonan syndrome, stated to Stat News that the result given by a genetic test is definite and more useful than these algorithmic testing results in identifying the particular genetic mutations.
“It’s inconceivable to me that one wouldn’t send off the panel testing and figure out which one it actually is,” told Gelb. However, he told that the algorithmic tests were “impressive”.
Further, he noticed that the dataset that was used for developing and testing the DeepGestalt algorithm was limited for fairly young children. Therefore the system could not be able to recognize the disorders in older persons easily, because when a person gets older, facial characteristics get less noticeable. A racial bias also has been suggested by the third party research done using FDNA’s tools.-Efficiency of these algorithms on Caucasian faces are higher than on African faces.
FDNA is having the knowledge of these limitations. Due to this, they introduce the DeepGestalt as “a reference tool”. That means this software assists in the diagnostic process but does not replace the human diagnoses.
Christoffer Nellåker, who is an expert in this field at the University of Oxford, gave this judgement to “New Scientist”.
“The real value here is that for some of these ultra-rare diseases, the process of diagnosis can be many, many years […] For some diseases, it will cut down the time to diagnosis drastically. For others, it could perhaps add a means of finding other people with the disease and, in turn, help find new treatments or cures.”