**7. Conclusions**

The availability of mobile phones to monitor patients with PD can have a profound impact on clinical practice, giving doctors access to long-term data. These additional data can help doctors obtain a more complete and objective understanding of the symptoms and fluctuations of their patients' symptoms and, therefore, will allow for more accurate diagnoses and treatment regimens.

We are creating a system of information and technological support for research on Parkinson's disease, taking into account the collection and processing of data of a large volume. Currently, in this system we have already created most of the modules, such as collecting data on the nature of the use of the phone and automatically filling out the patient's diary according to the data entered by the patient on the phone. Partially, we described this in Section 3 of this article. A more complete description can be found in one of our previous articles [22]. Using this system, data is sent daily to the doctor's computer. This eliminates the disadvantage of the lack of constant communication with the specialist, which we discussed in Section 2. The doctor has the opportunity to see a continuous graph of the

dynamics of the patient's condition. We have shown that it is possible to use neural networks, which can make filling out a patient's diary without his participation and make this filling out better.

As a result of this work, it was determined that the neural network is able to summarize data on the rotation angles and tilt angles of the mobile phone, in order to classify the condition of patients with PD. In this paper, we considered and carried out a comparative analysis of various architectures for building neural networks. An architecture of a neural network with two recurrent layers was chosen, at which an acceptable level of accuracy in classifying the state of a patient with PD was achieved.

Python neural networks were built using the TensorFlow and Keras libraries. It was established that the use of LSTM blocks instead of GRU leads to greater accuracy of the neural network. Future research will focus on the construction of neural networks for more parameters and a larger sample of the source data, which will lead to the training of the neural network so that it will return even more accurate diagnosis results.

**Author Contributions:** Conceptualization, Y.S.; methodology, Y.S., Y.I.; software, E.S.; validation, E.S., Y.S. and Y.I.; formal analysis, E.S.; writing—original draft preparation, E.S., Y.S.; writing—review and editing, Y.S. and Y.I.; visualization, E.S.; supervision, Y.I.; project administration, Y.S.; funding acquisition, Y.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** The research was funded by Russian Foundation for Basic Research (RFBR) according to the research project #18-57-34001.

**Conflicts of Interest:** The authors declare no conflict of interest.
