From 1500 diseases which are known to the science there are 150 which occur more often. The Skinive.ru have already taught a neural network to identify almost 30 types of neoplasms. They have had one year for it. Meanwhile, a human has to spend 3-4 years in a medical university to get such skill. World Health Organization claims that 22% of all people suffer from skin diseases.
Skinive it is not a diagnostic method neither doctors replacement. Technology creators identified their brainchild as a screening test. Now artificial intelligence is continuing collecting data. Just imagine that in 150 the algorithm will be able to recognize accurately each of 150 diseases. The founders of the project, Kirill Atstarov, and Yuri Evstratchik, told us why skiniving moles is good and being a skeptic is bad.
Legal telemedicine and pampering with technology
We created a telemedical marketplace in 2010. It was a platform where people can get an online consultation with a doctor. The project had been developing, we had passed acceleration in Moscow at the Internet Initiatives Development Fund. Then we asked ourselves if our activity was legal. It can be said, that we were one of those who promoted telemedicine in Russia. In 2017, Russian Federation accepted a law which permits provide medical services through telecommunications technologies. Up to 2017, such services could only be informational.
Soon after the law had been accepted such services as Yandex. Health appeared. It was impossible to compete on equal terms with such big players. At the same time, our marketplace had been aggregating a huge amount of medical data, and we wanted to implement it somehow. We had been developing different artificial intelligence solutions for 2 years, it was like pampering with the technology. We were in search of ways to develop our project. We had also been producing options for analysis of X-ray, MRI, mammography, ultrasound, etc. But we had faced the reality again: Belarusian healthcare system didn’t need such a technology, and as an application for ordinary people it was pointless.
There is hardly a person who owns an X-ray at home.
We stumbled upon research by Stanford University of melanoma detecting. This subject interested us, and we made our method. By the way, there are lots of applications for melanoma detecting, but there is any to detect basiliomas, carcinomas and other types of skin cancer. The point that all rivals offer binary classification — it is either melanoma or not melanoma. We increased the amount of skin diseases classes. Skinive is a unique project of its kind.
Later we left oncology and referred to viral skin diseases. Now we are teaching artificial intelligence to detect acne. There 6 main types of acne and only one can be treated with cosmetics. The other cases must be cured with medicines. Our priority task is to teach AI, the second is to teach users to take care of not outer beauty, but of health.
How it works
We all distinguish cats and dogs. At first glance, the features are the same: both cats and dogs have 4 paws, tails, fur. Our parents taught us to differentiate them, the showed us many times, that “it is a cat and it is a dog”. The process of a neural network teaching is the same. We added the technology of computer vision so our algorithm sees certain signs of skin neoplasms. As a result, we created a library of images that are classified by etymology. The entire library is passed through the algorithm, respectively, it already owns this data. Then we launch production to enable people to use the application. The more you submit correctly labeled data, the more accurately the neural network finds signs associated with a particular disease.
When we take a photo of a healthy mole, AI sees even edges, uniform color, small size, etc. and gives the result: "95% is a healthy mole." If there are signs of several diseases, we immediately send the user to a doctor.
You need to understand that Skinive is not an examination, not a diagnosis, it is a scoring test.
We are faced with the indignation of the doctors, they say: “Here you make the diagnosis, and who will be responsible for this?” And we explain to them that the diagnosis can only be made by the attending physician, we still do not claim it. Our task is to teach people to pay attention to tumors on their body. The sooner a disease is detected, the more likely it is that the treatment will end safely.
We may certify our software later to enable the application to make a diagnosis. We would like if AI will be useful for doctors in the future.
Why doctors shouldn’t be afraid of Skinive
On the one hand, doctors are overloaded. And on the other hand, 80% of patients of dermatologist-oncologist come with healthy moles, they don’t need to remove anything at all — that is, they simply create a queue. The third problem is that people with serious skin diseases often do not reach the doctor on time. They think that it will pass by itself.
A girl from our team found a flat wart on her body — this is a viral disease. If one such wart appeared, then the second and third will appear. Another team member also found a molluscum contagiosum, and molluscum contagiosum love to grow.
A guy from the North Caucasus once wrote to our tech support. These mollusks appeared just below his stomach. The guy writes: “Doctor, what is this?”. I answer him: “This is a molluscum contagiosum, we must go to the doctor — he will remove everything from you.”
The guy continues: “And what kind of ointment to smear?”.
I explain to him: “This can not be treated with ointment, it must be removed.” But after two weeks the same user writes again: “Doctor, these things are growing and increasing in number.” I explain again: “I told you that it will be so — you need to go to the doctor. Ointments will not help. " He: “Yes? Okay, got it. ” So a person, first of all, must make sure that there is a problem, and then he will get up and go to an appointment with a specialist. Skinive is just made for such a purpose.
By the way, our team has a person who interacts with doctors. Taking all the necessary doctors to the staff is too expensive. I told you, we had created the telemedicine marketplace — contacts of more than 1,200 doctors were collected there, of which 200 were involved in skin diseases. We have selected about 10 doctors with whom we consult on dermatology issues.
“A doctor will not work for free, but AI will”
Now we are negotiating with medical centers, as well as with developers of mobile applications in healthcare. We do want to go beyond one product. We have a special API service so any developer can integrate skin diagnostic functions using our technology in his mobile application.
Recently we had a meeting with one company which owns three medical centers in Minsk. They are already starting to use similar diagnostic solutions. It turned out that our service is much cheaper, more convenient to use, and it can recognize more diseases. We are ready for white label integration, for example, to place a chat-bot on the site of the medical center. Then each person will be able to take a photo, get a primary diagnosis and immediately make an appointment with a doctor.
We have ambitious development plans. We communicate with smartphone manufacturers to create our medical device. It will be equipped with medical programs and the camera, which will be able to shoot and do high-quality analysis. It will be a medical smartphone for skin diseases detecting.
The "Mole Map" function is currently under development on a 3D model of the human body. Imagine you took a picture of each mole, and the software saved the data. Then, for example, after a vacation in a hot country, you again check the moles, and the neural network gives you the result: one of the formations has changed, increased or the color has changed. This data might be a reason to attend a doctor.
We also plan to add to the application shortly a store with recommendations of medicines and an online appointment with a doctor. There are also plans to implement an online consultation with real doctors to verify the diagnosis. But, of course, the user will have to pay for the consultation. Because the doctor will not work for free. AI will be, but the doctor will not.
It’s easy to attract small financing in Belarus, but earning is hard
Last year we got some investments from business angels. Now we are in the second investments round, working on getting grants and negotiating with large investors. We must continue developing. So much data has been collected that our team can no longer cope with it. We need a bigger staff and to build a sales department. In general, in Belarus, it is easy to start and to attract the first small financing, but earning is hard. And finding large investors is also difficult.
How do we plan to earn? We want to set a price of 10 cents of one image analysis. This is for medical centers as soon as our software will help them to increase the number of target clients, so both sides will be in plus. As for ordinary users, we still plan to leave the application free for them. We will earn on the surrounding services. Perhaps the paid service “doctor’s opinion” will appear. Also, for each appointment to the doctor through our application, we will take a commission from medical centers.
Launching a startup is stress. It’s hard for us to reach out to doctors. Those doctors who work with us understand everything. And when you go out and try to interest someone in your technology, you are faced with skepticism and an absolute reluctance to even try to use the service. There is a category of doctors in Belarus who perceive AI as a competitor. We are trying to convey that Skinive is just a loyal assistant that can facilitate diagnosis and increase the flow of customers.