The new neural network is capable of analyzing electric activity in a brain which can be measured by an electroencephalograph or electrodes implanted directly into nervous tissue.
Alexander Khramov, Director of the Neuroscience Lab at the Innopolis University, and his colleagues used a direct distribution neural network in their experiments, which is considered to be a rather simple type of network. Data is transmitted in only one direction, from input neurons to output neurons, which allows simplifying the structure of AI and accelerating calculation.
The scientists explained that brain and its sections have structures and behavior similar to stochastic (random) systems – for example, such “unpredictable” phenomena as climate changes or stock prices. These systems are addressed by chaos theory in mathematics. Based on these ideas, the scientists trained AI to find similar patterns in brain activity signals.
After testing their AI on theoretical models, the neurophysiologists tested it on six epileptic rats. During seizures, the neural network analyzed data from the rats’ brain cortexes and areas deeper inside the brains that are presumably related to developing epilepsy. The AI helped to discover connections between different brain sections at rest and during seizures. Signals from one specific brain sections can be used to predict how another section may change. The scientists hope their research will help with studying brain disorders and in other scientific areas such as climate studies where scientists have to analyze data coming from several sources simultaneously.
Other existing medical neural networks include networks that can find traces of melanoma (skin cancer), breast cancer and signs of other diseases.