Artificial intelligence makes personalised stroke therapy possible

In Germany alone, around 270,000 people suffer a stroke each year. Doctors are under enormous pressure in their choice of therapy. A computer model should now assist them and make custom-fit decisions possible.

Röntgenbilder vom Hirn

DedMityay -

Following a stroke, patients need to be treated quickly. Every second counts. The main aim of therapy is to rescue as much brain tissue as possible. Doctors attempt to dissolve the blood clot that has caused the stroke with blood thinners or the use of a catheter, since the longer parts of the brain are deprived of sufficient blood and nutrients, the more tissue will die off. The consequences can be serious, ranging from loss of the ability to speak to symptoms of paralysis. But if the stroke occurred more than six hours ago, these therapeutic measures are no longer used, for medical professionals then estimate the risks of treatment as being greater than the benefits. Dr. Dietmar Frey is convinced that individual particularities of the affected patients are not sufficiently taken into account in making a decision based purely on the time factor. With his research team at the Charité – Universitätsmedizin Berlin, the neurosurgeon has developed a computer model based on artificial intelligence (AI) which assists doctors in choosing a therapy and takes greater account of individual factors.

“The generalised consideration deprives many patients of therapy options which can save their lives and make a decisive improvement to their quality of life,” explains Frey. “In certain cases”, he says, “a therapy may be useful that in other cases may be pointless after just three hours.” But doctors in the emergency life-saving station usually have little time for a complex case assessment. This is where the new computer model by the Berlin research team is intended to come in and take over the investigative work in the background.

Computer model trained with thousands of cases

The researchers have used machine learning for their computer tool. That means that they have trained the model with disease progressions and clinical data from thousands of stroke patients. On this basis it can now search in a very short time for comparable data sets and identify patterns which make a mathematically-based diagnosis possible for each individual case.

In future, it could look like this: straight after admission to hospital all relevant clinical patient data such as age, previous illnesses, blood pressure and laboratory results are analysed in the background by the computer model using comparable data sets. Meanwhile, MRT or CT scans are made of the patient’s brain and then these are included in the computer analysis as well. Twenty minutes later, when the doctors have to make their decision about therapy, the AI has already evaluated the information. “Then the doctor gets a personalised report of the results identifying the best possible treatment for each individual patient,” says Frey.

AI does not replace the doctor’s decision

The researchers are currently developing a user-friendly interface for their application together with a partner, which is intended to run on Tablets for day-to-day clinical use. For this, they are receiving funding of just under EUR 1.8 million from the German Federal Ministry of Education and Research (BMBF) under the GO-Bio programme. The formation of a spin-off is planned for the coming year, and the new software system is then set to prove itself in a comparative trial. In addition, the computer program is to continue to be trained and optimised constantly with new patient data. It is especially important for the neurosurgeon to stress that the computer is not the new “Doctor Data” that makes everything better. “We can’t replace doctors,” says Frey, “but simply help them to find the best possible therapy.”

Source: Newsletter „Aktuelle Ergebnisse der Gesundheitsforschung" Nr. 99 (in German)