With Chaos: A neural network trained to detect Alzheimer’s diseaseScience and technology18:17 | 20 June 2023

Researchers from Petrozavodsk State University, together with colleagues from India and Iran, have developed a neural network-based algorithm that can recognize symptoms of Alzheimer’s disease in electroencephalogram data with more than 70% accuracy, using chaos assessment. The scientists’ work has been published in the journal Algorithms.

The term “entropy” is used to evaluate the degree of chaos. Low values ​​indicate systematic, regular data, high values ​​indicate randomness. Various formulas are used to evaluate this. One of the most common is Shannon’s formula, in which a high entropy value is associated with the “unexpected” occurrence of an event.

In their study, the scientists proposed a completely different approach, using artificial intelligence to calculate entropy. To do this, they created a neural network that determines the entropy index NNetEn (Neural Network Entropy). The program is trained by converting the numbers to the time series given to it. As a result, the algorithm calculated the entropy value using mathematical methods.

The researchers demonstrated the benefits of the new method by using a machine learning algorithm to recognize Alzheimer’s disease in a ready database of electroencephalograms. Since the randomness of this signal changes in the presence of a disease, it was necessary to distinguish between healthy participants and patients based on the entropy value received by the neural network.

As a result, when using NNetEn as the sole feature, the accuracy of group separation was not very high – about 67%. However, the combination of a number of other entropies produced the desired result. When adding even an entropy paired with NNetEn, the accuracy of group separation increased to 73%.

Any researcher working with big data can easily apply the algorithm to their own data – the researchers emphasize that the neural network is in the public domain.