MedCFVQA：一种缓解医学视觉问答中模态偏好偏差的因果方法
MedCFVQA: A Causal Approach to Mitigate Modality Preference Bias in Medical Visual Question Answering
AI总结：
针对医学视觉问答中的模态偏好偏差，提出基于因果图的反事实模型MedCFVQA，并重构数据集以消除先验依赖，实验验证其有效性。
摘要：
医学视觉问答（MedVQA）对于提高临床诊断效率至关重要，它能就医学图像向临床医生的询问提供准确及时的答复。现有的MedVQA模型存在模态偏好偏差问题，即预测结果严重受单一模态主导，而忽略另一模态（在MedVQA中，通常是问题主导答案，而图像被忽略），从而未能学习多模态知识。为克服模态偏好偏差，我们提出了医学反事实VQA（MedCFVQA）模型，该模型使用偏差进行训练，并利用因果图在推理过程中消除模态偏好偏差。现有的MedVQA数据集在问题与答案之间表现出显著的先验依赖，这导致即使模型严重遭受模态偏好偏差，也能获得可接受的性能。为解决此问题，我们利用现有MedVQA数据集重建了新数据集，并在训练集和测试集中改变了问题与其答案之间的先验依赖（CP）。大量实验表明，MedCFVQA在SLAKE、RadVQA以及SLAKE-CP、RadVQA-CP数据集上均显著优于其非因果对应模型。
英文摘要：
Medical Visual Question Answering (MedVQA) is crucial for enhancing the efficiency of clinical diagnosis by providing accurate and timely responses to clinicians' inquiries regarding medical images. Existing MedVQA models suffered from modality preference bias, where predictions are heavily dominated by one modality while overlooking the other (in MedVQA, usually questions dominate the answer but images are overlooked), thereby failing to learn multimodal knowledge. To overcome the modality preference bias, we proposed a Medical CounterFactual VQA (MedCFVQA) model, which trains with bias and leverages causal graphs to eliminate the modality preference bias during inference. Existing MedVQA datasets exhibit substantial prior dependencies between questions and answers, which results in acceptable performance even if the model significantly suffers from the modality preference bias. To address this issue, we reconstructed new datasets by leveraging existing MedVQA datasets and Changed their P3rior dependencies (CP) between questions and their answers in the training and test set. Extensive experiments demonstrate that MedCFVQA significantly outperforms its non-causal counterpart on both SLAKE, RadVQA and SLAKE-CP, RadVQA-CP datasets.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
