基于多目标优化的医学影像可解释性增强研究OA

To address the need for reliable interpretability in medical imaging,this study proposes a multi-objective particle swarm optimiza-tion-enhanced explanation framework that improves explanation quality and clinical readability by optimizing the LIME(Local and Model-Ag-nostic Explanations)process.The proposed method incorporates a multi-objective search strategy into the LIME pipeline,enabling an adaptive trade-off between explanatory fidelity and regional sparsity,and producing Pareto-optimal explanation outcomes.Experiments conducted on knee X-ray images from a publicly available knee osteoarthritis dataset using representative convolutional neural networks demonstrate that the method increases fidelity by up to 18%and reduces sparsity by up to 22%,resulting in more focused and stable explanations.These results in-dicate that the proposed framework offers a feasible and effective pathway toward trustworthy AI-driven medical image interpretation.