基于混合受限玻尔兹曼机的肺结节良恶性诊断

发布时间：2018-03-28 04:13

本文选题：受限玻尔兹曼机　切入点：肺结节　出处：《计算机工程与应用》2017年23期

【摘要】：针对传统计算机辅助诊断中肺结节的特征提取方法依靠人工设计、操作复杂、识别率低等问题,提出了一种基于混合受限玻尔兹曼机的肺结节良恶性诊断方法。首先采用多层无监督卷积受限玻尔兹曼机自动对肺结节图像进行特征学习,然后利用分类受限玻尔兹曼机对获得的特征进行良恶性分类。为避免分类受限玻尔兹曼机在训练中出现的特征同质化问题,引入了交叉熵稀疏惩罚对其进行优化。实验结果表明,该方法有效避免了手动特征提取的复杂性,在肺结节良恶性分类的准确率、敏感性、特异性、ROC曲线下面积值上均优于传统诊断方法。
[Abstract]:In view of the traditional computer-aided diagnosis of pulmonary nodules feature extraction methods rely on artificial design, complex operation, low recognition rate and other problems. A new method for the diagnosis of benign and malignant pulmonary nodules based on mixed constrained Boltzmann machine is proposed. Firstly, multi-layer unsupervised convolution constrained Boltzmann machine is used to automatically study the features of pulmonary nodules. Then the classification constrained Boltzmann machine is used to classify the obtained features. In order to avoid the problem of homogeneity in the training of the classification constrained Boltzmann machine, Crossover entropy sparse penalty is introduced to optimize it. Experimental results show that the proposed method can effectively avoid the complexity of manual feature extraction and is accurate and sensitive in the classification of benign and malignant lung nodules. The area under the specific ROC curve was superior to the traditional diagnostic method.
【作者单位】：太原理工大学计算机科学与技术学院;山西省人民医院;
【基金】：国家自然科学基金(No.61540007,No.61373100) 国家重点实验室开放基金(No.BUAA-VR-15KF02,No.BUAA-VR-16KF13)
【分类号】：R734.2;TP391.41

【相似文献】