Dual Causal Inference Framework for Medical VQA

A new framework called Dual Causal Inference (DCI) has been introduced by researchers to enhance Medical Visual Question Answering (MedVQA) by addressing cross-modal confounding effects. DCI merges Backdoor Adjustment (BDA) and Instrumental Variable (IV) learning to manage both seen and unseen confounders. BDA is employed to reduce observable biases, such as common visual-textual co-occurrences, while IV learning addresses the unobserved confounders. This innovative framework is founded on a Structural Causal Model (SCM). Notably, it is the first comprehensive architecture that integrates these two causal inference techniques for MedVQA, with the goal of bolstering the reliability of diagnostic reasoning.