News: Medical AI in Healthcare
No matter how pessimistic the skeptics are — artificial intelligence is increasingly being used in the fields of health and medicine. Today, the machine learning technology can effectively predict the development of diseases, search for new drugs, diagnose diseases by MRI/CT or x-ray images. Let’s take a closer look at the most promising developments in this field.
Tasks that are not directly related to the treatment of patients
Modern systems of artificial intelligence already help doctors to treat patients. For example, HeartFlow algorithm is able to build a 3D map of the heart using CT scans and deep learning technology. This provides an opportunity for doctors to diagnose heart diseases more accurately and quickly, reducing the number of invasive procedures required by 80%.
- HeartFlow creates a personalized, digital 3D model of the arteries.
- Powerful computer algorithms solve millions of complex equations to assess the impact that blockages have on blood flow.
- The result is a color-coded map that aids clinicians in determining, vessel-by-vessel, if sufficient blood is reaching the heart.
According to the project team HeartFlow-guided strategy reduced the overall costs to the healthcare system by more than $4,000 per patient after one year. The mean one-year per-patient cost for the usual care strategy was $12,145 compared to the $8,127 cost for the HeartFlow-guided strategy. When including the $1,500 cost of the HeartFlow Analysis, the cost reduction is 26 percent.
Improving the emergency departments
Artificial intelligence systems and machine learning can cope with more than just diagnosis. For example, at the end of May, the University College London Hospital in Bloomsbury (UCLH) announced that it would use AI systems to identify patients who really need emergency medical care.
When the patient arrives to the emergency room, complaining of some kind of pain or injury, the medical staff performs standard procedures — blood analysis, collects anamnesis, if necessary, makes an x-ray. As noted in the clinic, in 80% of cases there is nothing serious the patients just receive a prescription for medicine and can safely go home.
The artificial intelligence system will make it possible to quickly identify those 20% who really need emergency care.
Searching for new knowledge
Treatment practices followed by doctors tend to become obsolete. New methodologies, new research and drugs are emerging. Back in 2004, researchers studied the content of 341 medical journals and found that the total number of monthly publications exceeded 7000.
Ideally, the physician needs to constantly maintain the level of subject knowledge, to be aware of the modern practices of treatment, however, to examine the entire body of publications that are regularly published in thematic magazines, is almost impossible — even if we are talking about a narrow specialist.
Artificial intelligence technologies can help in this situation. Such a solution was developed by scientists from the American research center - RAND, engaged in methods of analysis of strategic problems. They taught the system to search for key words and terms related to the subject of the request in huge amounts of information.
During the tests, this topic was data on gout, low bone density and osteoarthritis of the knee joint. The algorithm was able to reduce the number of relevant articles of interest to doctors by 67–83%. According to the developers, the system missed only two articles that would have been selected by people, but none of them contained critical information. The accuracy of the machine learning algorithm was 96%.
Creating new drugs
The experience of pharmaceutical companies shows that it takes about 12 years from pre-clinical trials to drug approval. At the same time, only 0.1% of the “candidate drugs” get to clinical tests. Approval is received by only 20% of them.
Artificial intelligence systems can help resolve this situation and accelerate the release of new drugs. Machine learning and AI systems are used in the early stages of drug development.
For example — a system called AtomNet uses deep learning techniques to predict how molecules will behave and how likely they will form the necessary bonds.
During the training, the developers of AtomNet uploaded into the system large amounts of data on the results of several million already known interactions of molecules. Based on these interactions, the system has learned to predict interactions that have not yet occurred. The PA has already helped to develop Ebola drugs.