A neural network for diagnosis of Parkinson's disease according to the Archimedes spiral.

Abstract

The article considers the topical problem of diagnosing Parkinson’s disease in the early stages of the disease.  It is stated that there is no diagnosis for exact determination of the diagnosis now, so it remains to carry out tests to detect symptoms.  The choice of the study topic is due to the incurability of Parkinson's disease, therefore, detection of the disease in its early stages, according to the authors, is extremely urgent. One of them, as the authors rightly note, is Archimedes spiral drawing, which has a rather high accuracy in detecting resting tremor. Creating an algorithm for automatic processing of such images can help in the diagnosis and monitoring of disease progression. The combination of artificial intelligence and Internet medical things will eventually make connected health monitoring devices more intelligent. Neural networks and the vast amounts of data generated by Internet medical things could also be used to make diagnoses. The authors of the article overlay the fundamental capabilities of neural networks and their favorable role in transforming the field of radiology, by saving time and money for medical organizations. A neural network has been developed that will be able to make a diagnosis based on the image of the drawn Archimedes spiral. This network can be used for early diagnosis and further monitoring of the disease. Due to a small sample of images for training and training of the model, the authors of the article decided to increase the sample by image transformation, as well as the use of convolutional neural network with pre-training. As a result of this work, a model with 93.7 percent accuracy was created, which will allow to automate the process of diagnosing the disease in its early stages.