Uncertainty-guided graph attention network for parapneumonic effusion diagnosis

•This is the first work to distinguish and classify UPPE, CPPE and normal cases using a deep learning method.•We propose an uncertainty-guided graph attention network (UG-GAT) to capture and represent the spatial information and rich contextual information of the given 3D volumetric data, so as to improve the classification performance.•We propose uncertainty measurement for the guidance of the model in order to reduce data noise and uncertainty, and further concentrate on the most salient slices and symptoms, which help the network to bypass the requirement for a large-scale annotated dataset.