A tree-structure-guided graph convolutional network with contrastive learning for the assessment of parkinsonian hand movements

•A tree-structure-guided graph convolutional network with contrastive learning scheme is developed for automated and objective video assessment on Parkinsonian hand movements.•A novel tri-directional skeleton tree scheme is developed for effective fine-grained extraction of spatial features.•A tree max-pooling module is designed to improve the learning ability of the model to salient fine-grained motion features.•A group-sparsity-induced momentum contrast is developed to capture more discriminative spatial-temporal dynamics and realize stable feature learning.