Is Medical AI the next Breakthrough in Healthcare?
In 2017 and 2018 AI has become a large industry, yet in previous years it was mostly considered as just a technology. Stand-alone cases of AI implementation in some key industries emerged into continuous integration of AI into programs for digital products and technological processes.
The most popular technological trends are recommendation systems based on deep learning algorithms, natural language understanding (NLU) or natural language interpretation (NLI) systems, computer vision, predictive modelling and reinforcement learning.
There are several types of companies, which operate in the AI market:
- Research laboratories, which create algorithms;
- Companies (most likely — startups), which dream of an exponential growth of their products thanks to the integration of artificial intelligence;
- Consulting agencies, helping business to understand its needs and integrate new technologies, as well as offering b2b solutions.
The number of companies, operating in the field of AI development, grows constantly. For example, the number of active AI startups in 2018 grew twofold in comparison to the year 2015 as well as their funding has also doubled.
There are around thirty-five times more vacancies, which require knowledge in the field of deep neural networks development nowadays in comparison to 2015. In recent years, even some special government policies emerged regarding AI.
The leaders in the field of R&D (research and development) are internal laboratories of Google, Amazon, Microsoft and Netflix and some commercial organizations (OpenAI, Vector Institute).
Medicine — the Most Promising Sphere for AI Implementation
As for today, medical AI is one of the most popular technologies for investors. Since 2013, medical AI startups managed to attract $4,3 billion in 576 transactions (according to CB Insights). In three years the medical AI market will reach $6,6 billion, increasing every year by 40%. There is nothing strange about it — this technology is to revolutionize almost all medical spheres in the nearest years.
Almost all leading digital corporations are now developing AI-based medical products and services. According to the Venture Scanner research company, there are more than 800 companies worldwide, which are involved in this process.
What will happen?
The thing is that AI can fundamentally change the way medical diagnostics is performed (even at home), it can affect the development of new drugs, not to mention the possible increase in overall quality of medical services and lowering the expenses for medical clinics.
Why it should happen?
At this moment, diagnostics mistakes, such as misdiagnosis and delayed diagnosis, are one of the most significant problems in modern healthcare. In general, medical errors are the third leading cause of premature mortality after CVDs and cancer. Unfortunately, it has always been a common case when doctors are misdiagnosing and are prescribing wrong treatment after detecting symptoms of completely other disease. According to British Medical Journal 2016 report, misdiagnosis is responsible for more than 250000 deaths in the U.S. Other sources (e.g. James, John T. PhD article for Journal of Patient Safety) suggest even bigger figures, evaluating the annual mortality up to 440000.
According to the Federal Compulsory Medical Insurance Fund, Russian doctors are mistaken with treatment in 10% of cases. As Lev Kaktursky (the main pathologist of the Ministry of Health of Russia) notes, the discrepancy between the posthumous and lifelong diagnoses tends to be 20–25%. Therefore, the fourth part of the deaths of patients in the Russian Federation comes from a disease that was not found, when the patient was alive.
In addition, the problem of the lack of qualified medical practitioners is widely present in developing countries. For example, in India, a country with more than 1.324 billion population, the doctor — patient ratio is evaluated to 1:1674, which is significantly higher than World Health Organization norm of 1:1000. It is estimated that the country needs 500000 more physicians to fill the deficit, which will be extremely difficult to do considering the demographic growth rate. In poor countries of Sub-Saharan Africa, the situation is even much worse. Particularly in Tanzania, the ratio is about 1:50000, Weill Cornell Medical College.
The most recent developments in the field of medical AI can help to solve these problems, as well as a wide number of others.
High Risks and Possible Problems
One of the biggest problems that impedes the start of the full-scale integration of AI-assisted diagnostics into healthcare is the lack of required medical data. Datasets, the large sets of medical images, which are used for the process of neural network training, must be labeled. It means that a qualified medical specialist, which makes the process very time-consuming and costly, must manually mark out any disease or abnormality on medical images. Today there is no specific demand for the labeled datasets from the developers, which makes the development of the whole sector stall.
Considering the current AI technology development, there is also a high risk of possible monopoly. Large corporations, such as IBM and Google, are seeking to bring their own products to the medical neural network market, monopolize the market and continue to receive the majority of revenue. If this happens, it will certainly increase the cost of such services and will negatively affect their quality.
Considering the current AI technology development, there is also a high risk of possible monopoly. Large corporations, such as IBM and Google, are seeking to bring their own products to the medical neural network market, monopolize the market and continue to receive the majority of revenue. If this happens, it will certainly increase the cost of such services and will negatively affect their quality.
The Solution
In order to prevent the monopolization of the medical AI market we should pay attention to the projects with distributed structure. One of them is Skychain, which is now extremely close to the moment, when the platform is fully operational.
Skychain intends to provide an infrastructure to radically increase the efficiency of healthcare AI development and training. It will make diagnostic AI systems far more accessible and affordable for the consumer by using blockchain technologies to facilitate safe transactions between the key parties.
The project is based on the revolutionary approach of using smart contracts to bring together many healthcare big data providers whose data is vital for AI training, thousands of independent AI developers, computational resource providers, and millions of consumers.
All of the features, listed above, are already implemented and working in the system. More than that — six neural networks are already available for public testing in Skychain alpha.
These AI’s can already diagnose a various number of diseases. To show the power of the infrastructure, the developers decided to focus on the most common, though, most dangerous ones:
- Respiratory diseases (atelectasis, cardiomegaly, effusion, infiltration, mass, nodule, pneumonia, pneumothorax, consolidation, edema, emphysema, fibrosis, pleural thickening, hernia). Average accuracy — 84%.
- Bone abnormalities recognition (detects fractures, degenerative diseases, lesions and subluxations). Average accuracy: 82%.
- Skin cancer detection. Average accuracy — 70%.
- Brain tumors detection. Average accuracy — 87%.
- Breast cancer detection. Average accuracy — 93%.
- Retinal diseases (DME, CNV and drusen) detection. Average accuracy — 99%.
When looking at these numbers a question arises — why some of the ANN’s have only 70% accuracy, when others have almost 100%? The fact is that this problem has already been mentioned above and it is the lack of marked medical datasets, needed to be “fed” to the neural network in order to increase its accuracy. Today this exact problem should be solved first, and Skychain has found a way to do it.
The key problem for lack of necessary training data is because medical institutions simply do not want to share it with independent developers as long as they do not receive any revenue from this cooperation.
Setting up these symbiotic relationships between medical institution laboratories with large datasets and independent developers, who are looking for medical datasets, which could be used to train their AI’s, is the key to solving this problem, and Skychain provides a simple yet brilliant mechanism for such purposes.
Skychain for medical data providers is a data marketplace where they can provide their data as a service for training third-party neural networks.
Trained neural networks on the supplier data sets can only be used with Skychain, so the data set that providers will receive includes a guaranteed share of revenue from all neural networks trained, based on their data.
The preparation of medical data sets will be beneficial for market participants who own this data. Skychain takes care of the intellectual property of developers of neural networks and data owners. A developed and trained neural network, as well as the data set with which it was trained, are protected from outside participants.
To sum up: patients, doctors, medical institutions and online services will be able to use the Skychain capabilities by paying for each use of a neural network with special tokens and creators of these ANNs along with the medical institutions, which provided data for the training process, will be rewarded for each use of the artificial intelligence.
The beta version of Skychain is avaliable right now at https://beta.skychain.global/
