摘要
前列腺癌是男性恶性肿瘤中发病率最高的肿瘤之一.通过对前列腺癌的有效诊断，可以尽早治疗，降低前列腺癌的死亡率.针对前列腺癌早期筛查的准确度不足，病理检查给病人身体带来严重负担的现状，以及现有基于数据挖掘的癌症诊断方法只关注诊断结果准确性或者只关注可解释性的问题，本文提出一种基于多目标神经网络的诊断方法，通过特征选择提取对诊断结果最具有解释性的特征子集，以提高模型的可解释性和准确度；通过采用进化计算的方式进行神经网络的结构和权重学习，从而构建有效的能够充分体现临床信息与前列腺癌之间关联的多目标神经网络模型进行前列腺癌诊断；并通过Pareto优化方法对模型训练过程中的结构和参数进行优化，从而提供多个有效的诊断模型以满足医务工作者不同的决策偏好.
Prostate cancer is one of the highest incidence of cancer in male. The most effective way to reduce prostate cancer mortality and treat patients is to detect it earlier. So far, the accuracy of early screening of prostate cancer is still unsatisfactory and pathological examinations seriously hurt patients body, as well as the existing cancer diagnosis method based on data mining is only focus on the accuracy or interpretability of diagnostic results. According to these problems, this paper proposes a multi-objective neural network-based diagnostic model. In our approach, feature selection is carried out to extract the most explanatory subset of features, thereby improving the explanatory capability and accuracy of the model. Evolutionary computation is employed to learn the network structure and weights, with which the correlation between clinical information and prostate cancer can be identified for diagnosis of prostate cancer. And the Pareto optimization method is used to optimize the structure and parameters of the model during training process, thus providing a set of effective diagnostic model to meet the different decision-making preferences of medical workers.
出处
《系统工程理论与实践》
EI
CSSCI
CSCD
北大核心
2018年第2期532-544,共13页
Systems Engineering-Theory & Practice
基金
国家自然科学基金(71533001)
中央高校基本科研业务费专项资金资助项目(DUT15QY32)~~
关键词
前列腺癌诊断
多目标神经网络学习
进化计算
特征选择
prostate cancer diagnosis
multi-objective neural network learning
evolutionary computation
feature selection
