G2GNet: A model for video tremor diagnosis in Parkinson’s disease

Parkinson’s disease (PD) diagnosis constitutes a fundamental and formidable research area in the realm of intelligent healthcare. Tremor, widely regarded as a hallmark manifestation of PD, plays a pivotal role in the diagnostic process. Consequently, investigations focusing on tremor in the context of PD hold tremendous value for accurate disease diagnosis and assessment. In this study, we introduce a novel deep learning architecture, G2GNet, which diagnoses parkinsonian tremor by analyzing video data. Capitalizing on the strengths of Graph Convolutional Networks (GCN) and OpenPose, the model employs cross-entropy loss function and L2 regularization to enhance its diagnostic classification accuracy, thereby streamlining parkinsonian tremor detection. Experimental evidence demonstrates that G2GNet outperforms existing methodologies when evaluated on the Tim-Tremor dataset, achieving an impressive accuracy rate of 94.0%. This advancement offers a potent supplementary tool for the diagnosis and evaluation of tremor symptoms in Parkinson’s disease patients.