CausalDisenSeg: A Causality-Guided Disentanglement Framework with Counterfactual Reasoning for Robust Brain Tumor Segmentation Under Missing Modalities

2026-04-14

AI Summary

This work addresses the vulnerability of existing multimodal brain tumor segmentation methods to style-induced biases when MRI modalities are missing, which often leads models to rely on spurious correlations and compromises robustness. To tackle this issue, the study introduces structural causal models into the task for the first time, proposing a causally guided disentanglement framework. This framework explicitly separates anatomical causal factors from style bias factors via a conditional variational autoencoder and integrates a region-wise causal module with dual adversarial counterfactual reasoning. Through a three-stage causal intervention pipeline, the method achieves debiased segmentation. Evaluated on BraTS 2020, it significantly outperforms current approaches, maintaining high accuracy even under severe modality缺失. Cross-dataset evaluation on BraTS 2023 yields a macro-averaged Dice Similarity Coefficient of 84.49, demonstrating strong generalization and robustness.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

In clinical practice, the robustness of deep learning models for multimodal brain tumor segmentation is severely compromised by incomplete MRI data. This vulnerability stems primarily from modality bias, where models exploit spurious correlations as shortcuts rather than learning true anatomical structures. Existing feature fusion methods fail to fundamentally eliminate this dependency. To address this, we propose CausalDisenSeg, a novel Structural Causal Model (SCM)-grounded framework that achieves robust segmentation via causality-guided disentanglement and counterfactual reasoning. We reframe the problem as isolating the anatomical Causal Factor from the stylistic Bias Factor. Our framework implements a three-stage causal intervention: (1) Explicit Causal Disentanglement: A Conditional Variational Autoencoder (CVAE) coupled with an HSIC constraint mathematically enforces statistical orthogonality between anatomical and style features. (2) Causal Representation Reinforcement: A Region Causality Module (RCM) explicitly grounds causal features in physical tumor regions. (3) Counterfactual Reasoning: A dual-adversarial strategy actively suppresses the residual Natural Direct Effect (NDE) of the bias, forcing its spatial attention to be mutually exclusive from the causal path. Extensive experiments on the BraTS 2020 dataset demonstrate that CausalDisenSeg significantly outperforms state-of-the-art methods in accuracy and consistency across severe missing-modality scenarios. Furthermore, cross-dataset evaluation on BraTS 2023 under the same protocol yields a state-of-the-art macro-average DSC of 84.49.

Problem

• missing modalities
• modality bias
• brain tumor segmentation
• spurious correlations
• robustness

Innovation

• Causal Disentanglement
• Counterfactual Reasoning
• Structural Causal Model
• Modality Robustness
• Brain Tumor Segmentation