Application of machine learning in neurophysiological data analysis (AI-Med)

Parkinson’s disease affects between 0.15% and 0.3% of the general population. The percentage increases with age – up to 1.5% of the over-65 population has Parkinson’s disease. In Poland, the prevalence of Parkinson’s disease is 180 people per 100,000 residents. There are more than 10 million people living with Parkinson’s disease worldwide and another case is diagnosed every minute.

The treatment is usually based on pharmacotherapy. When medications are not effective, treatment may involve implanting electrodes in the brain. Those electrodes, by generating impulses, modify the action of the nerve centers responsible for the symptoms of the disease and ailments are significantly reduced. Incorrect placement of the probe can lead to serious consequences – from lack of treatment effects through emotional disorders to brain damage, which can eventually lead to limb paresis or loss of vision. Our project addresses this very challenge – where to place the probe to minimize the risk of complications or rule them out altogether.

Working with the National Research Institute of Oncology and the patients who have agreed to help us, we have developed a breakthrough solution – a system based on neural networks that, by “listening” to brain activity, helps the doctor decide exactly where to place the electrode so that the treatment is as effective and safe for the patient as possible. From a 30-second audio recording, the algorithm selects the clearest half-second and, based on this, determines precisely where to place the probe and shows this in an operator-friendly, graphical form.

For the past four years, our neural networks, together with a team of neurosurgeons at the National Research Institute of Oncology, have aided the recovery of dozens of people per year. With each operation performed, our system is getting increasingly perfect because it learns from each new case.

The Deep Brain Stimulation (DBS) method was first used in 1948. Since then, the number of applications of this method in neurosurgery has steadily increased. Our solution may be useful for other DBS surgeries. The network can “listen” to other areas of the brain and help treat diseases that cause symptoms similar to Parkinson’s Disease – tremors in the limbs or involuntary spasms. An example of such a disease is dystonia.

In addition, in medicine, research is underway to use the DBS method to treat psychiatric disorders, especially drug-resistant depression. The brain is still a very poorly understood organ, and algorithms capable of eavesdropping and analyzing its activity – will find more and more applications.

You can read more about our project in the Mathworks article.