Sergey Nikolenko – Chief Research Officer, Neuromation
These days, Artificial Intelligence has been establishing itself in most diverse areas ranging from Go game to unmanned ground vehicles. Below are presented three directions in contemporary research which employ latest AI achievements (primarily, deep neural networks) for biomedical applications. This particular research may both transform and substantially extend human life.
Polar, Beiersdorf AG and others. Smart clothing
There is a range of projects which will hardly rock the whole world, but will certainly make adjustments to the daily life in the immediate future. That is AI application in Internet of Things (IoT), meaning bodily things.
All sorts of fitness trackers and special bracelets which collect information on pulse, steps, etc. have become commonplace. For the sportswear manufacturers the main trend of today is incorporating various sensors in the cloths. This way, the collected information will be more diverse and accurate. Sensors for smart garment were invented in 2016, and in 2017 Polar presented its Polar Team Pro Shirt which monitors a lot of information during sports exercises. Still, the main thrill will come later when sports medicine employs AI to properly use the information obtained. That will be astonishing even for Moneyball.
And the process has been launched. Last November 24-26, SkinHack hackfest took place in Moscow. It was about applying machine learning models in analysis of data gathered by such a pro-shirt. The first SkinHack was arranged a year earlier and devoted to recognizing and analyzing imperfections of human skin, when its participants attempted to establish the age of an individual based on the skin structure and wrinkles seen on photos. Smart cosmetics and smart textiles are of prime interest for Beiersdorf AG (commonly known as Nivea), therefore appearance of the said technologies in the market will hopefully not take long. In Russia, SkinHack was supported by Youth Laboratories affiliated with InSilico.
InSilico: new cure machine research
InSilico Medicine has been well reputed in biomedicine for a long while. Its principal mission is inhibiting ageing. Hopefully, InSilico will succeed in this undertaking. Its other somewhat related but nevertheless differing project is automated search of new medicines.
A medicine is a chemical mixture capable of interconnecting with other agents of a human body (primarily, proteins) in order to improve some required features. They can suppress some proteins while stimulating generation of others at a higher scale. Creating a new medicine means learning to select from a huge variety of options of chemical combinations those which will produce a required impact.
Fully automated medical research is impossible since clinical test, tests on mice and then humans are inevitable. A process of taking a new medicine to the market usually takes long years. But doctors may be assisted if the research period is shorter. InSilico develops machine learning models which attempt to not just prognosticate the properties of certain compounds, but to generate molecules of preset propensities, thus helping to choose the most hopeful candidates for laboratory and clinical tests.
That is accomplished through a most interesting deep learning model, generative adversarial networks (GAN). These networks include two components. A generator strives to synthesize new objects, molecules of required properties, for example, whereas a discriminator attempts to distinguish the generated information from original data. While learning to deceive a discriminator, a generator thus starts to generate objects which are undistinguishable from the original ones. We do hope they will be original. The latest InSilico model was named druGAN (drug + GAN), and it is attempting to synthesize, among other things, molecules which can cure oncology diseases.
MonBaby baby care
Finally, there is a project which Neuromation intends to join. Young children, babies first of all, who can not ask for help themselves, require special attention and care. Such attention is at times required even in situations when Mom and Dad could seemingly relax, when their baby is asleep, for instance, but can move and hurt a leg. Sudden Infant Death Syndrome (SIDS) should not be forgotten either, since its risk is very much dependent on the position of a baby when asleep. SIDS risk is growing multifold if a baby is asleep face down.
MonBaby smart baby monitor is a small smart button snapped onto baby’s outfit to monitor breathing and movement of the sleeping baby. The application is currently based on machine learning to analyze timing series. The data about baby’s movements is used to distinguish breathing cycles and detect if the baby is sleeping on his or her stomach or back.
Neuromation intends to supplement the system with smart cameras able to monitor baby’s movements via images. For that, it will use computer vision systems based on deep convolution networks, and synthetic data for their learning. The problem is, collecting sufficient volume of original data for the system learning is in fact impossible. That would require video recordings of dozens of thousands of babies including recordings of all sorts of critical situations. Fortunately, the present-day medical practices and human ethics just do not permit generating such data. Neuromation therefore plans to create virtual babies, 3D models allowing to simulate required situations and generate artificial video imaging for learning.
So all three above developments in various biomedicine areas – sports medicine and cosmetics, creating new medications and caring for babies – all actively employ the latest AI achievements, but those are just few examples. AI is now used in hundreds of diverse biomedical projects. All that is an illustration of AI’s ability to make human life better, healthier and longer.