This research proposes a novel solution for medical image classification using Graph Neural Networks (GNNs). The approach uses Image Foresting Transform to segment images, which are then transformed into graph-structured data for feature extraction. An ensemble of three distinct GNN architectures is used to boost robustness. The proposed method surpasses traditional Deep Neural Networks (DNNs) in performance while reducing the parameter count. This reduces costs, accelerates training, and minimizes bias, making it a viable and scalable strategy for medical image classification with less dependency on extensive training datasets.
Publication date: 14 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.07321
