AI can now be used to identify genetic diseases
According to a recent study, it has been revealed that a new artificial intelligence (AI) technology can accurately identify some of the rarest genetic disorders from a patient’s photograph.
This technology, DeepGestalt performed better than some of the most respected clinicians in identifying a range of syndromes in three trials. Additionally, it could also add considerable value in personalized care, as quoted by the study (https://www.nature.com/articles/s41591-018-0279-0) published this Monday in the journal of Nature Medicine.
The study further claims that 8% of the entire population has diseases with key genetic components which may be indicated by recognizable facial features.
For example, this technology can successfully identify Angelman Syndrome – a disorder of the nervous system that is characterized by facial features like a wide mouth, widely spaced teeth, strabismus (where the eyes point in different directions), or a protruding tongue.
In a conversation with CNN, Yaron Gurovich, the Chief Technology Officer at FDNA said, “It demonstrates how one can successfully apply state of the art algorithms, such as deep learning, to a challenging field where the available data is small, unbalanced in terms of available patients per condition, and where the need to support a large amount of conditions is great,” FDNA is an artificial intelligence and precision medicine company and Gurovich was the one leading the research.
He also confessed that this can prove to be a major breakthrough as it would open up a whole new world of opportunities for future research and applications and even in identifying new genetic syndromes or diseases!
However, the authors of the study also warned about the potential downside to the technology because facial images are easily accessible and future employers may easily find out about these conditions through this software and discriminate against suffering individuals.
The deep learning algorithm DeepGestalt was trained by the entire team led by Gurovich, by using 17,000 facial images of known patients from a database. This database contained patients who were diagnosed with more than 200 different genetic diseases.
The team discovered that the AI technology performed better than established clinicians in two different sets of tests to identify a particular syndrome in 502 selected images. For every test, the AI proposed a list of potential syndromes and identified the right syndrome from among the top 10 suggestions 91% of the time.
In case of Noonan syndrome, the algorithm succeeded in identifying it 64% of the time while in a previous study, the clinicians could only do so in 20% of the cases by looking at the images.
The technology applies the deep learning algorithm to the facial characteristics in the image before coming up with a list of possible diseases or syndromes.
Though it does not cite the facial features that led to the prediction, but it produces a heat map visualization to help the researchers understand the exact regions on the face which helped the technology classify the syndromes.
Though it does look promising, but the performance of the AI system is tough to measure in absence of any public benchmarks to compare it to. While there still are limitations to the technology and there is a lot of tweaking still needed to make sure that the algorithm is robust in the real world hospital environment, but it is clear that there is immense potential when it comes to AI in the healthcare system.