Mccseg: Morphological Guided Causal Constraint Network for Medical Image Multi-Object Segmentation

Abstract

Medical image multi-object segmentation aims to accurately extract each object that is great significance for the medical image analysis. Although several methods based on deep leaning have been widely developed, it is still challenging due to the similarities of different objects, the significant variability of morphology, data heterogeneity and the unbalanced distribution of samples. To address the above issues, we propose a novel Morphological Guided Causal Constraint segmentation network (MCCSeg) for medical image multi-object segmentation. We build a hybrid CNN-Transformer encoder module to obtain local and global features. Specifically, a Causal Constraint Module (CCM) is proposed for feature decorrelation by sample reweighting, which utilizes weight allocation to train the weight of training samples. The Random Fourier Features (RFF) are used to solve nonlinear dependency problems and remove correlations between features for eliminating the impact of spurious correlations. The morphological guidance module (MG) is designed to extract the boundary as prior morphology information for enhancing feature representation, which are concatenated into the deep supervision module in decoder for further optimizing feature extraction. The experiments demonstrate that MCCSeg outperforms other state-of-the-art methods, obtaining up 3.76% and 5.41% improvements in DICE and HD95 scores on Synapse dataset, respectively. And the ablation experiments further verify the effectiveness of MCCSeg in improving generalization. The source code will be released at https://github.com/YfGAO/MCCSeg.

Keywords: medical image, multi-object segmentation, deep learning, causal learning, morphological guidance