Fine-grained strabismus diagnosis method and system using graph neural network based on causal feature selection, and device and medium

Abstract

A fine-grained strabismus diagnosis method and system using a graph neural network based on causal feature selection, and a device and a medium. The method comprises: acquiring nine facial photographs of a patient, and performing image preprocessing on each of the nine facial photographs, in order to construct a nine-gaze-position image; using an object detection algorithm to detect an orbital region in each gaze position in the nine-gaze-position image, in order to extract feature variables of each gaze position that are related to fine-grained strabismus diagnosis; using a causal feature selection algorithm to select from among the feature variables related to each gaze position key feature variables having a direct causal relationship with fine-grained strabismus diagnosis; and inputting the key feature variables of each gaze position into a graph neural convolutional network model, and performing training and optimization with a strabismus disease dataset by means of a propagation formula, in order to output a fine-grained diagnosis result. Unrelated details are removed from entire facial images, and main eye regions are extracted to construct a nine-gaze-position image, thereby reducing the computational cost; and a causal feature algorithm is used to extract key feature variables of each gaze position, and a graph neural convolutional network model is used to learn the key feature variables of each gaze position, thereby making the graph neural convolutional network model more interpretable, and thus realizing fine-grained strabismus diagnosis.

Images