TY - JOUR
T1 - CausalMixNet
T2 - A mixed-attention framework for causal intervention in robust medical image diagnosis
AU - Zhang, Yajie
AU - Huang, Yu An
AU - Hu, Yao
AU - Liu, Rui
AU - Wu, Jibin
AU - Huang, Zhi An
AU - Tan, Kay Chen
N1 - Publisher Copyright:
© 2025 Elsevier B.V.
PY - 2025/7
Y1 - 2025/7
N2 - Confounding factors inherent in medical images can significantly impact the causal exploration capabilities of deep learning models, resulting in compromised accuracy and diminished generalization performance. In this paper, we present an innovative methodology named CausalMixNet that employs query-mixed intra-attention and key&value-mixed inter-attention to probe causal relationships between input images and labels. For mitigating unobservable confounding factors, CausalMixNet integrates the non-local reasoning module (NLRM) and the key&value-mixed inter-attention (KVMIA) to conduct a front-door adjustment strategy. Furthermore, CausalMixNet incorporates a patch-masked ranking module (PMRM) and query-mixed intra-attention (QMIA) to enhance mediator learning, thereby facilitating causal intervention. The patch mixing mechanism applied to query/(key&value) features within QMIA and KVMIA specifically targets lesion-related feature enhancement and the inference of average causal effect inference. CausalMixNet consistently outperforms existing methods, achieving superior accuracy and F1-scores across in-domain and out-of-domain scenarios on multiple datasets, with an average improvement of 3% over the closest competitor. Demonstrating robustness against noise, gender bias, and attribute bias, CausalMixNet excels in handling unobservable confounders, maintaining stable performance even in challenging conditions.
AB - Confounding factors inherent in medical images can significantly impact the causal exploration capabilities of deep learning models, resulting in compromised accuracy and diminished generalization performance. In this paper, we present an innovative methodology named CausalMixNet that employs query-mixed intra-attention and key&value-mixed inter-attention to probe causal relationships between input images and labels. For mitigating unobservable confounding factors, CausalMixNet integrates the non-local reasoning module (NLRM) and the key&value-mixed inter-attention (KVMIA) to conduct a front-door adjustment strategy. Furthermore, CausalMixNet incorporates a patch-masked ranking module (PMRM) and query-mixed intra-attention (QMIA) to enhance mediator learning, thereby facilitating causal intervention. The patch mixing mechanism applied to query/(key&value) features within QMIA and KVMIA specifically targets lesion-related feature enhancement and the inference of average causal effect inference. CausalMixNet consistently outperforms existing methods, achieving superior accuracy and F1-scores across in-domain and out-of-domain scenarios on multiple datasets, with an average improvement of 3% over the closest competitor. Demonstrating robustness against noise, gender bias, and attribute bias, CausalMixNet excels in handling unobservable confounders, maintaining stable performance even in challenging conditions.
KW - Front-door adjustment
KW - Medical image diagnosis
KW - Patch mixing
KW - Unobserved confounders
UR - http://www.scopus.com/inward/record.url?scp=105004733949&partnerID=8YFLogxK
U2 - 10.1016/j.media.2025.103581
DO - 10.1016/j.media.2025.103581
M3 - 文章
AN - SCOPUS:105004733949
SN - 1361-8415
VL - 103
JO - Medical Image Analysis
JF - Medical Image Analysis
M1 - 103581
ER -