Welsh poet Dylan Thomas passionately called for a struggle against the inevitability of death. The leading futurists of our time echo echo his calls. The only difference in sentimentality between the poet and prosaicness reality. We all gonna die someday. The only question is when and how.
Or not.
Currently, scientists trying to make artificial intelligence, machine learning and computer vision to predict the death of a person. The ultimate goal is, of course, is to make the AI the grim Reaper, and in time to prevent chronic diseases and other illnesses.
Recent studies of this application of AI in medicine used the latest machine learning model to analyze CT scan 48 breast cells. The computer was able to predict which patients will die within five years, with 69% accuracy. About the same as you would have predicted any doctor.
The results were published in the scientific reports journal Nature authored by teams from the University of Adelaide. Lead author Dr. Luke Ogden-Rainer, radiologist and PhD student, says one of the obvious advantages of using AI in precision medicine is the early detection of health risks and potential intervention.
Less obvious is the promise to accelerate research on longevity.
“Currently, most studies of chronic diseases and longevity require long periods of observation that will allow us to identify the differences between patients with and without treatment because these diseases are slowly progressing,” he explains. “If we could quantitatively assess changes before, not only were we able to identify the disease, but also to implement more effective intervention, and would also be able to react much earlier”.
This can lead to faster and cheaper treatment. “If we could knock off a year or two of time that is required for the transfer of therapeutic agents from the laboratory to the patient, the progress in this field was greatly accelerated.”
AI heart
In January, researchers from Imperial College London published results that showed that AI can predict heart failure and death better than a human doctor. The study, published in the journal Radiology, involved the creation of a virtual three-dimensional hearts 250 patients that can mimic cardiac function. Then, the algorithms of AI have been studying what functions will serve as best predictors. This system relied on the MRI, blood tests, and other data.
In the end it turned out that the car was faster and better in determining the risk of pulmonary hypertension — showed 73% accuracy against 60%.
Scientists say that this technology could be used to predict outcomes in other cardiovascular diseases in the future. “We would like to develop a technology that could be used for a variety of cardiovascular diseases to help the physician interpret the results of medical tests,” says study co-author Tim Davies. “The goal is to see if the best projections to facilitate proper treatment and longevity of the people.”
The AI gets smarter
This kind of use of AI in the field of precision medicine will only get better as the machine will be trained like medical students.
Ogden-Rainer said that his team continues to create the perfect data set as it moves forward, but increased predictive accuracy from 75 to 80 percent, include such information as age and gender.
“I think there is an upper limit to how accurate we can be, because there will always be an element of randomness,” he says, answering the question how well the AI will determine the mortality of the individual. “But we can be more accurate than now, if we take into account the risks and power of individuals. A model combining all these factors, hopefully, will be able to clarify the risk of short-term mortality up to 80 percent.”
Other more optimistic look at how quickly the AI converts this aspect of medicine.
“Prediction of remaining life for people, in fact, one of the most simple applications of machine learning,” said Dr. Ziad Obermeyer. “It requires a unique set of data that is contained in electronic records-related information about the time of death. As soon as we collect enough such data, we can very accurately predict the likelihood that an individual will live a month or a year.”
AI is still learning
Experts like Obermeyer and Oakdene-Rainer agree that progress is coming quickly, but there is still much work to be done.
On the one hand, there is still much data that you can dig, but they are not yet ordered. For example, images in which machines are trained, have still to handle, making them useful. “Many groups of scientists around the world spend millions of dollars on this task, because it remains a bottleneck to successful medical AI”, say Oakden-Rainer.
In an interview with the News STAT Obermeyer said that data is fragmented across the healthcare system, so linking information and creating datasets will take time and money. He also notes that though there is great excitement on the subject of using AI in precision medicine in a clinical setting, these algorithms almost have not been tested.
“You can say that all is well and the algorithm is really good. But now you need to bring it into the real world and to see what happens,” he says.
AI — not an accident
To prevent a deadly disease is one thing. But is it possible to prevent fatal case with AI?
It is intended to do American and Indian scientists when they attended to the growing number of deaths among people doing a selfie. The group identified 127 people who died posing for a photo, for two years.
Based on the combination of text, images and location, the machine learned to identify the self as potentially dangerous or not. Mileage at more than 3,000 signed selfies on Twitter showed 73% accuracy.
“The combination of features from the image and place showed the best accuracy,” say the researchers.
What’s next? Will appear a system warning for selfie lovers.
AI and the future
All this discussion gave rise to the question: do we really know when we die?
According to one study, recently published in Psychology Review, the answer is no. 9 out of 10 people in Germany and Spain, when asked whether they want to know about your future, including death, have preferred to remain ignorant.
Obermeyer looks at this question differently: through the lens of people who live with life-threatening illness.
“Among the things that patients really want and what you don’t get is answers from doctors on the question “how long have I got?”. Doctors are very reluctant to answer these questions, partly because I don’t want to be wrong about important things. Partly because the patients themselves do not want to know.”
Can artificial intelligence predict death?
Ilya Hel