Understanding how technology helps doctors save millions of lives
– how artificial intelligence can facilitate the work of doctors
– what is wrong with computed tomography of people’s hearts
– why people do not trust new technologies in medicine
MedTech or medical technology is an extremely promising area for both large technology corporations and little-known startups. Although the latter very often become the engine of innovation in the market.
E if it weren’t for modern technology in medicine, we would still have to hold arsenic in our mouths and shove intimidating devices of different sizes and shapes.
What can I say, back in the middle of the last century, people in our country were tortured by cutting tonsils, and children could have rave parties using quartz procedures.
Fortunately, today even a smart watch can draw your cardiogram and warn about heart problems.
Not to mention more advanced devices, like bio-implants to replace human bones and organs, as well as smartphone apps that can detect symptoms of breast cancer.
Even robots have learned to perform complex surgical operations. But much more important is the development of systems for the diagnosis and prevention of fatal diseases.
Occupational doctors sometimes may not notice serious symptoms, which often leads to untimely treatment and death.
That’s why MedTech developers have high hopes for a variety of artificial intelligence technologies. (AI). Neural networks in medicine make it possible to process huge patient databases and identify specific diseases in advance much more quickly and efficiently.
Many companies and startups are already working on systems that will allow doctors to predict the condition of patients and warn them of possible threats. One of these neural networks is being prepared by the Ukrainian startup Cardio Vision.
AI guard your heart
According to the World Health Organization, nearly 20 million people die from cardiovascular diseases each year. This represents one third of the annual deaths on the planet and makes heart disease the most common cause of death.
The most common among these diseases is coronary heart disease, which leads to heart attack, angina pectoris, cardiac arrhythmias, and heart failure. A third of all deaths from these diseases occur in people under 70 years of age.
Recently, the Cardio Vision team, which was founded by students of the Ukrainian Catholic University (UKU) in Lviv, has developed an AI-based web service to combat coronary heart disease.
In 2019, the idea of the Cardio Vision service won the international competition Microsoft AI for Good Challenge, and now the Ukrainian neural network is being prepared for clinical tests in Australian hospitals, which is where Cardio Vision customer comes from.
As told HB startup founders Bogdan Petrishak, Marichka Dobko and Oles Dobosevich, their product is modern software for hospitals, counseling centers and any medical institutions that have computed tomography (CT).
“In coronary heart disease, atherosclerotic plaques grow on the walls of the coronary artery, which cause narrowing of the artery. This affects the blood supply to the myocardial heart muscle and greatly increases the risk of fatal diseases, ”Cardio Vision explains.
Lviv startup plans to introduce an AI system in hospitals that will allow doctors to upload CT images of coronary angiography of patients and determine the degree of occlusion of coronary arteries (stenosis level).
The problem is that the CT coronary angiography procedure generates a very large amount of data, and experts simply do not have time to analyze them. Cardio Vision says that doctors spend about half of their work time interpreting CT images of the heart.
In addition, even after analyzing the data, doctors are not always able to determine which of the patients needs emergency care.
The new Ukrainian neural network will allow uploading CT coronary angiography images to a web service, where the algorithms will automatically analyze all the data and predict the level of stenosis. Depending on the condition of the coronary tree, the AI will divide the prognosis into three categories – a healthy area, an insignificant level of stenosis or a dangerous stage of blockage.
“The doctor receives predictions of the general condition of the coronary artery and can instantly determine the stage of the disease of a particular patient. By analyzing all patients who underwent CT coronary angiography, the service will help identify those who need emergency medical care in the first place and inform the doctor about it, ”said the founders of the neural network from UCU.
Cardio Vision also allows specialists to test the operation of AI and find out why the algorithm assigns certain images to one of three categories. To do this, it is enough to open a specific image and highlight the areas that the neural network allocates in connection with various blockages and narrowing of the coronary arteries.
Such a system does not replace the doctor, but only supplements it, making routine work many times faster and more efficient.
When will the Ukrainian neural network be ready and who are its competitors?
Now there is only a pilot version of the main Cardio Vision product, but in the next six months, project managers plan to test their algorithms in real conditions.
The security of patient data also presents additional difficulties with such software – the Ukrainian team will need to ensure confidential processing of information in accordance with the legislation of the country in which they are going to work. Only after that they will be able to enter the global market.
“In parallel with the development of the project in Australia, we would be very happy if we could develop in Ukraine. Unfortunately, now in Ukraine there is not enough equipment that would generate the necessary data array, ”Cardio Vision explains.
Among its competitors, the startup highlights TeraRecon, Cohesic, Circle and other companies that automate the analysis of heart condition data.
But the Ukrainian team has some advantages: access to a huge database of patients, which allows you to train neural network algorithms; fully automatic analysis of the level of stenosis in CT images of coronary angiography. Therefore, Cardio Vision may become unique.
Although the implementation of this system is worth the haste, because every day in the field of MedTech new technologies appear.
For example, in early February 2020, doctors from University College London announced the development of AI, which can accurately measure the blood flow of the body, determining on the basis of this the risk of heart attack and the probability of death.
Algorithm developers used images of cardiovascular magnetic resonance (CMR) more than a thousand patients, and after studying them with a neural network, the prognosis of the system was compared with the assessment of qualified doctors.
“AI is moving from computer labs to the real world of healthcare, performing some tasks better than doctors alone. We tried to measure blood flow manually before, but it is tiring and time-consuming, ”said Professor James Moon of University College London.
The neural network helped doctors discover that patients with reduced blood flow had an increased risk of death due to heart attack, stroke, and heart failure.
According to Moon, such predictions ultimately lead to better patient care, and also give doctors a new understanding of how the heart works.
However, AI in medicine is only at the initial stage of development. In the near future, such systems should allow doctors to build a causal relationship between different symptoms and diseases, instantly analyzing patient data.
The future of neural networks in medicine
Most computer systems now allow only the correlation between the many conditions of patients and their diseases to be determined. This does not always provide useful information for doctors during treatment.
In the past few years, AI developers have used algorithms that can detect specific causes of diseases and their symptoms in a single database.
Now experts are preparing such neural networks that will allow the integration of existing databases, which should seriously enhance the accuracy of the analysis.
In November 2019, researchers at the British company Babylon Health, which provides healthcare services, introduced a computer system to combine different patient databases.
According to Anish Dhir and Siaran Li, Babylon Health developers, their product will allow you to use large amounts of unused medical data to find out the causes and consequences of diseases, and, possibly, discover new cause-effect relationships.
Dhir and Lee created a chatbot application that can diagnose users and offer treatment after they fill out a special form and indicate their symptoms.
Such a service can save time for both patients and doctors, as it immediately eliminates people who do not need to see a doctor.
The authors of the application said that their AI can diagnose the state of the human body better than doctors. Because of this, some doctors have criticized the Babylon Health product, claiming that the neural network does not see the symptoms of some serious diseases.
However, the technology was checked by experts, and last week it was presented at the conference of the Association for the Development of AI (AAAI) in New York. This organization was founded in 1979, is one of the most respected in the field of AI, and now it includes more than 4000 representatives around the world.
“The justification of the answers, which indicates the main cause and effect, rather than hidden correlations, should give people more confidence in the application. Health is a high-risk area. We don’t want to provide a black box, ”says one of the creators of the application, Siaran Lee, who also studies quantum programming at University College London.
Unlike the typical AI machine learning method, Babylon Health used quantum cryptography. Such mathematical formulas protect the processed data from outside interference.
The system was tested using databases in which cause-effect relationships were already established – by analyzing the size and texture of breast tumors, the AI determined whether they were malignant or benign.
The authors of the application are sure that if “raw data ”will be available, their algorithm will be able to determine the relationship between symptoms and disease in the same way as clinical studies.
Siaran Lee says the U.S. Food and Drug Administration is approving new drugs based on trials that show only correlation.
In contrast, their algorithm may be more efficient and help professionals already at the initial stage. But, for this you need to instill confidence in people to use such systems.