Replication Data for: GraphScrib: Integrating Graph Neural Networks with CNN-Transformer for Medical Image Segmentation
Description
This repository contains the supplementary dataset, sparse scribble annotations, data splits, and pre-trained model weights associated with the paper "GraphScrib: Integrating Graph Neural Networks with CNN-Transformer for Medical Image Segmentation". The research introduces a novel weakly-supervised semantic segmentation (WSSS) framework that unifies Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), and Graph Neural Networks (GNNs) to address the partial activation anomaly in medical imaging. Contents of this repository: Scribble Annotations: The generated sparse scribble masks used as weak supervision signals. Data Splits: The exact Train/Validation/Test patient splits used for the ACDC and MSCMRseg datasets to ensure fair comparison and reproducibility. Pre-trained Weights: The final model weights for GraphScrib achieving the reported Dice scores. Note: The raw MRI images belong to the original ACDC and MSCMRseg challenges and should be obtained from their official platforms.