Welcome to the future: Are hospitals ready for Artificial Intelligence?
Artificial intelligence is already proving useful for medical tasks that are time-consuming and boring. But in the future AI might suggest a course of action for a patient, determine if a tumour is benign or aggressive, and even offer a second opinion.
According to Dietmar Frey, a neurosurgeon at Berlin’s Charite, one of the largest university hospitals in Europe, the technology could soon be used to treat stroke patients.
“This is more that just an idea. We have the technology, and a prototype,” Frey says.
The project began in May under his supervision with a team of doctors and IT experts. And he’s already convinced that it will work.
When someone is having a stroke, you need to respond fast. Time is of the essence: if the brain is not receiving enough blood and oxygen, parts of it slowly begin to die.
The consequences of not acting quickly can be disastrous: loss of speech and paralysis are common effects. If the patient doesn’t find their way to a hospital within four hours, it’s often too late.
“After that [doctors] don’t treat anymore,” says Frey. Generally, this is because any further treatment could do even more damage. But is that true for every patient?
Frey thinks that some patients probably still could respond well to treatment after the four-hour mark, whereas others can’t be helped even after two hours. But it’s difficult for a doctor to tell which is the case – and fast.
This is where researchers believe a computer could make all the difference. A machine could search through and compare thousands of archived cases within minutes and tell the doctor what can – or cannot – be done.
Some hospitals already use artificial intelligence to support doctors and nurses. AI can be especially useful for tasks that are time-consuming and boring, during which a human might lose concentration.
AI might sound scary if you’ve been reading and watching a lot of science fiction. But we’re not in the realm of Isaac Asimov or his laws of robotics quite yet.
Today, AI just means there is a computer that is compiling and searching through large amounts of data, and learning to recognize patterns.
“In simple terms, the computer recognizes: On this picture is a cat, on the next picture is a dog,” explains Benedikt Brors from the German Cancer Research Centre (DKFZ) in Heidelberg.
The computer can then recognize a cat and a dog every time you present that to it – but if you show the computer a horse, then it won’t know what that is. The more input the computer receives, the more it can recognize and the more accurate it becomes.
Over the past 10 years, the DKFZ has been working on an AI that can tell doctors if neuroblastoma – a rare kind of tumour that affects children – are benign or aggressive, as this significantly changes the treatment options.
But it will take time until clinical trials can be conducted, as the certification process is complicated.
Dietmar Frey’s computer currently has around 2,000 stroke cases in its system. The individuals have all been broken down into data points: age, gender, weight, smoking habits, pre-existing conditions, medication, image data and lab results. Frey’s plan is to have this data package up and running in the ER when a patient arrives at the hospital.
It takes between 10 and 15 minutes in the emergency room to diagnose a stroke, and another 10 to decide on which treatment is necessary.
Frey would like to have a computer begin comparing a patient’s case with the data set immediately after they have been registered at the hospital. The computer could then see what methods had been successful in the past, and suggest a course of action within a few minutes.
“A doctor doesn’t have the time or the capacity to start looking through old files for comparisons in the emergency room. And they can’t keep it all in their head,” says Frey.
If Frey’s clinical trial is successful, then the system could be brought to other clinics with the help of funding from the government or the EU.
Radiologist and computer scientist Felix Nensa of Essen University Hospital has been working on creating a database of images containing CT scans and x-rays of lung conditions.
The AI database allows doctors to highlight parts of a scan and have the computer compare it with previous cases, if they exist. Each image comes with a full medical background and a diagnosis.
“To me, this AI is the first wave of tools that lighten the doctors’ load,” Nensa says.
“In five to 10 years there will be more of these tools. The profession of doctors will change – more towards amalgamation and interpretation of data,” he speculates.
Physicist and pathologist Frederick Klauschen also agrees that AI will become an important part of diagnostics, as it will help doctors concentrate on the important questions. “A pathologist can do qualitative work better, a computer is better at quantitative work,” he says.
Maybe someday soon computers will be able to give doctors a second opinion on diagnosis and treatment – which some find scary.
“What makes it a bit creepy for many people is that some of these systems’ rules aren’t directly transparent,” says Brors. “An outsider can’t trace back why the computer has made a specific decision.”
This doesn’t really matter, as long as the computer makes reliably correct and good decisions. “But it makes some doctors sceptical. It’s like a black box. They can’t check if it is plausible anymore.”
What a machine can’t do yet, is be creative or act on instinct. If it is presented with an unusual case that doesn’t line up with the data, then it simply won’t know what to do.
Humans will probably always be part of the solution. Because although computers can and must be part of the future of medicine, there are some things machines can’t do. And if not humans, who else will be there to think outside the box?
Ulrike von Leszczynski is a writer and correspondent for DPA.