基于TPE-SVM的乳腺癌诊断可解释人工智能方法

Explainable AI Approach for Breast Cancer Diagnosis via TPE-Optimized Support Vector Machine

摘要: 乳腺癌是女性中最常见的恶性肿瘤之一，早期精准诊断对提高患者生存率至关重要。针对传统乳腺癌诊断方法存在的主观性强和误诊率高，以及现有人工智能模型可解释性较差等问题。本文设计一个基于贝叶斯优化支持向量机(TPE-SVM)的乳腺癌智能诊断模型，并结合LIME解释方法提高诊断过程中的可解释性。方法上，构建基于径向基核函数的支持向量机模型，并利用TPE算法对关键超参数进行优化，最后引入LIME方法实现诊断结果的特征可视化与解释。采用UCI数据库中的威斯康星乳腺癌数据集进行仿真验证，结果显示，该方法在各项指标上表现优异，进一步的LIME可解释性分析也表明，模型判别依据与临床医学知识高度一致。本文所构建的诊断框架为AI在医疗场景中的可用性与可信性提供了新思路。

Abstract: Breast cancer is one of the most common malignant tumors among women, and accurate early diagnosis is crucial for improving patient survival rates. To address the high subjectivity and misdiagnosis rates of traditional diagnostic methods, as well as the limited interpretability of current artificial intelligence models, this study proposes an intelligent diagnostic model for breast cancer based on a Bayesian-Optimized Support Vector Machine (TPE-SVM), integrated with the LIME explanation method to enhance interpretability during the diagnostic process. Methodologically, a Support Vector Machine with a radial basis function kernel is constructed, and its key hyperparameters are optimized using the TPE algorithm. The LIME method is subsequently employed to visualize and interpret the model’s diagnostic decisions. The model is validated on the Wisconsin Breast Cancer dataset from the UCI repository. Experimental results demonstrate excellent performance across multiple evaluation metrics, and further LIME-based interpretability analysis confirms that the model’s decision-making criteria are highly consistent with established clinical knowledge. The diagnostic framework proposed in this study offers new insights into the usability and trustworthiness of AI applications in medical scenarios.