ارزيابي شدت بيماري پاركينسون با استفاده از سنجنده كينكت

Parkinsonʹs disease is one of the most common group of progressive neurological disorders that affects millions of people worldwide and is characterized by movement problems such as stiffness, slowness, and tremors. Accurate and objective assessment of the severity of this disease plays an important role in monitoring its progress and eva‎luating the effectiveness of therapeutic interventions. In recent years, computer vision technologies have shown promising potential in providing non0invasive and quantitative measurements for the assessment of Parkinsonʹs disease. Today, it has been observed that precognition and a stage of Parkinsonʹs disease can be identified by the FAB objective test by measuring executive function impairment. FAB is a brief and accurate test that has a correlation between this test and other diagnostic tests for the aforementioned disease, such as the Unified Parkinsonʹs Rating Scale (UPDRS), but it is time0consuming. The severity of Parkinsonʹs disease is of particular importance in terms of predicting the stages of the disease and prescribing movements. Therefore, models have been designed that help the above tests. In this research, the goal is to create a system for determining the severity of Parkinsonʹs disease that takes into account the skeletal images obtained from the Kinect camera and uses deep learning to train the network. The Kinect camera, used as a depth0sensing device in the field of human0computer interaction, captures 3D skeletal data of people performing a series of standard movement tasks, and then uses this data using advanced learning techniques. The machine was processed to extract relevant features and quantify the severity of movement disorders associated with Parkinsonʹs. The proposed system uses the unique capabilities of the Kinect camera to accurately track and analyze key movement patterns such as gait, tremor, and bradykinesia, which are commonly observed in affected individuals, using the spatio0temporal graph convolutional network (STGCN) and adaptive graph convolutional network algorithms. Two0stream (2SAGCN), can identify specific movement abnormalities and provide objective measurements of disease severity. In this research, transfer learning has been used for network optimization and 3D data augmentation, which improved the results. The STGCN and 2SAGCN models were trained on the IRDS data set and with the addition of the channel attention module, they achieved test accuracy of 92.63 ± 0.32% and 93.71 ± 0.40% respectively in expressing the severity of Parkinsonʹs disease and the movement performed by the person.

6