The Welsh poet Dylan Thomas passionately called for the fight against the inevitability of death. The leading futurists of our time echo his calls. The only difference is the poet's sentimentality and the prosaic nature of reality. We all die someday. The only question is when and how.
Or not.
Scientists are currently trying to make artificial intelligence, machine learning and computer vision predict human death. The ultimate goal is, of course, not to turn AI into a grim reaper, but to prevent chronic diseases and other illnesses in time.
The latest research into this AI application in medicine has used the latest machine learning models to analyze CT scans of 48 chest cells. The computer was able to predict which patient would die within five years with 69 percent accuracy. Just like any doctor would predict.
The results were published in Nature Science Papers by a team at the University of Adelaide. Lead author Dr. Luc Oukden-Rainer, a radiologist and graduate student, says one of the clear benefits of using AI in precision medicine is the early identification of health risks and potential intervention.
Less obvious will be the promise to accelerate durability research.
“Currently, most studies on chronic disease and longevity require long follow-up periods to tell the difference between treated and untreated patients because these diseases progress slowly,” he explains. "If we could quantify changes earlier, we would not only be able to identify the disease, but we would be able to intervene more effectively, and we could react much earlier."
This can lead to faster and cheaper treatment. "If we could cut off a year or two of the time it takes to transfer drugs from the laboratory to the patient, progress in this area would be significantly accelerated."
Promotional video:
AI with heart
In January, researchers at Imperial College London published results that showed AI could predict heart failure and death better than a human doctor. The study, published in the journal Radiology, involved creating virtual 3D hearts of 250 patients that can mimic heart function. Then AI algorithms began to explore what functions would serve as the best predictors. This system relied on MRI, blood tests, and other data.
Ultimately, it turned out that the machine was faster and better at identifying the risk of pulmonary hypertension - 73% accuracy versus the usual 60%.
Scientists say the technology could be used to predict the outcomes of other cardiovascular diseases in the future. “We would like to develop technology that could be applied to a wide variety of cardiovascular diseases to help doctors interpret medical test results,” said study co-author Tim Daves. "The goal is to see if the best predictions can contribute to the right treatment and longevity in people."
AI is getting smarter
This kind of AI applications in precision medicine will only get better as machines learn like medical students.
Oakden-Reiner says his team continues to build the perfect dataset as they move forward, but has already improved predictive accuracy from 75 percent to 80 percent by including information such 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, when asked how well AI will determine the mortality of an individual. “But we can be more accurate than we are now if we take into account the risks and strengths of individuals. A model combining all these factors will hopefully be able to refine the risk of short-term mortality up to 80 percent.”
Others are even more optimistic about how quickly AI is transforming this aspect of the medical field.
“Predicting the remaining life for humans is actually one of the simplest uses of machine learning,” says Dr. Ziyad Obermeyer. “It requires a unique set of data, which is contained in electronic records associated with information about the time of death of a person. Once we collect enough of this data, we can predict very accurately the likelihood that an individual will live for a month or a year, for example."
AI is still learning
Experts like Obermeyer and Oakden-Rainer agree that progress is coming quickly, but there is still a lot of work to be done.
On the one hand, there is still a lot of data to dig into, but it is not ordered yet. For example, images that machines learn from still have to be processed to make them useful. “Many teams of scientists around the world are spending millions of dollars on this task because it remains a bottleneck for successful medical AI,” Oakden-Rainer says.
In an interview with STAT News, Obermeier said that data is fragmented throughout the healthcare system, so linking information and creating datasets will take time and money. He also notes that while there is a lot of hype about the use of AI in precision medicine, these algorithms have hardly been tested in the clinical setting.
“We can say that everything is fine and the algorithm is really good. But now we need to take him out into the real world and see what will happen with full responsibility,”he says.
AI is not an accident
Preventing fatal disease is one thing. But can a fatal accident be prevented by AI?
This is exactly what American and Indian scientists intended to do when they were concerned about the growing number of deaths among people taking selfies. The team identified 127 people who died posing for a photo over the course of two years.
Based on a combination of text, images, and location, the machine has learned to identify selfies as potentially dangerous or not. A run of over 3,000 signed Twitter selfies was 73 percent accurate.
"The combination of image and location based features showed the best accuracy," the scientists say.
What will happen next? A warning system for selfie lovers will appear.
AI and the future
This whole discussion has raised the question: Do we really want to know when we will die?
According to one of the papers recently published in Psychology Review, the answer is no. 9 out of 10 people in Germany and Spain, when asked if they wanted to know about their future, including death, chose to remain in the dark.
Obermeier looks at this issue differently: through the prism of people who live with a life-threatening illness.
“Among the things that patients really want and do not get, these are the answers from doctors to the question 'How much do I have left?' Doctors are very reluctant to answer these questions, in part because they don't want to be wrong about such important things. Partly because the patients themselves don't want to know."
Ilya Khel