基于集成SVM的肺部肿瘤PET/CT三模态计算机辅助诊断方法

本研究提出基于集成SVM的肺部肿瘤PET/CT三模态计算机辅助诊断新方法.首先在临床采集肺部肿瘤患者PET、CT和PET/CT各2000例三模态图像数据上提取对同一病灶ROI区域;然后根据CT、PET和PET/CT的不同特点,从三模态图像的ROI区域中提取形状特征、灰度特征、Tamura纹理特征和GLCM特征等不同特征分别构成80、98、98维特征分量,并分别在不同特征空间里构造个体分类器,包括CT-SVM、PET-SVM、PET/CT-SVM;最后,基于相对多数投票原则,对CT-SVM、PET-SVM和PET/CT-SVM进行集成,识别对肺部肿瘤.实验结果表明,该方法能够有效提高肺部肿瘤的诊断正确率.

SVM Ensemble based Computer-aided-diagnosis Method of Lung Tumor using PET/CT Multi-modality Data

To propose a new method that PET/CT computer-aided-diagnosis of lung tumor based on SVM ensemble.Firstly, 2 000 cases of PET, CT and PET/CT in patients with lung cancer from clinic were collected, and the ROI of the same lesion location for three modal images was extracted.Secondly, according to the different characteristics of CT, PET and PET/CT, shape feature, gray feature, Tamura and GLCM feature and other features were extracted from CT ,PET, PET/CT image ROI,and the 80,98 and 98 dimensional feature space were constitued and the individual classifier in different feature space was constructed, including CT-SVM, PET-SVM, PET/CT-SVM;Thirdly, CT-SVM,PET-SVM and PET/CT-SVM were ensemble to use for computer-aided-diagnosis for lung tumor.The experimental results show that this method can effectively improve the diagnostic accuracy of lung tumor.