MAMBO-NET: Multi-Causal Aware Modeling Backdoor-Intervention Optimization for Medical Image Segmentation Network

TL;DR

This work tackles bias in medical image segmentation caused by confusion factors through a causal backdoor framework. It introduces MAMBO-NET, which combines Gaussian Self-Modeling (GSm) to represent latent confounds and a Bias-Aware Causal Intervention (CIBM) to realize backdoor adjustment, guided by posterior constraints via KL loss and boundary-focused uncertainty loss. The approach explicitly models $P(Y^{\prime}|do(X))$ by learning distributions over confusion factors and fusing bias-mitigated features during decoding, achieving Dice gains up to $2.28\%$ on ultrasound datasets and reduced false-alarm rates on dermatoscopy and colonoscopy data. Across five datasets, MAMBO-NET demonstrates enhanced segmentation accuracy and robustness under varied imaging conditions, highlighting the importance of causal-interventional strategies and latent-confounder modeling in medical image analysis.

Abstract

Medical image segmentation methods generally assume that the process from medical image to segmentation is unbiased, and use neural networks to establish conditional probability models to complete the segmentation task. This assumption does not consider confusion factors, which can affect medical images, such as complex anatomical variations and imaging modality limitations. Confusion factors obfuscate the relevance and causality of medical image segmentation, leading to unsatisfactory segmentation results. To address this issue, we propose a multi-causal aware modeling backdoor-intervention optimization (MAMBO-NET) network for medical image segmentation. Drawing insights from causal inference, MAMBO-NET utilizes self-modeling with multi-Gaussian distributions to fit the confusion factors and introduce causal intervention into the segmentation process. Moreover, we design appropriate posterior probability constraints to effectively train the distributions of confusion factors. For the distributions to effectively guide the segmentation and mitigate and eliminate the Impact of confusion factors on the segmentation, we introduce classical backdoor intervention techniques and analyze their feasibility in the segmentation task. To evaluate the effectiveness of our approach, we conducted extensive experiments on five medical image datasets. The results demonstrate that our method significantly reduces the influence of confusion factors, leading to enhanced segmentation accuracy.

Figures (5)

• Figure 1: The segmentation results of multiple networks. The third row corresponds to a zoomed-in image of the selected area. The curves of different colors in the second and last rows represent the corresponding models' predictions of the lesion area.
• Figure 2: The architecture of the proposed MAMBO-NET. Dashed lines indicate that the data stream is disabled during model inference. Segmentation Encoder and Segmentation Decoder are the encoder and the decoder in UNeXt. Gaussian Backbone and Posterior Constrain Backbone will use the global average pooling(GAP) and linear mapping for scale alignment.
• Figure 3: The process of causal relationship modeling. X denotes the original image; Y denotes the predicted mask, and c denotes the confusion factors.
• Figure 4: Visualization results of segmentation of multiple models on ultrasound, dermatoscopy, and colonoscopy images.
• Figure 5: Feature entropy map generated by the decoder layer, where $K$ represents the number of distributions in the GSm.