基于多特征融合与XGBoost的肺结节检测

为了提高肺结节检测的精确度和效率，提出一种基于多特征融合和XGBoost的肺结节检测模型。首先采用阈值分割与形态学运算，获得候选结节区域；然后通过基于超分辨率重建的卷积神经网络进行候选结节的特征增强；其次采用快速鲁棒特征、灰度共生矩阵、灰度不变矩的提取方法获得候选结节的局部与全局的多种特征，采用词袋模型进行降维并融合；最后利用XGBoost-决策树分类模型去除假阳性结节，完成肺结节的检测。在LIDC-IDRI数据上进行的实验表明该模型能达到97.87%的准确率和97.92%的召回率。该模型可用于辅助医生进行肺结节诊断，具有一定的临床应用价值。

Lung nodule detection based on multi-feature fusion and XGBoost

Abstract: A lung nodule detection model based on multi-feature fusion and XGBoost is proposed for improving the accuracy and efficiency of lung nodule detection. After candidate nodule regions are obtained by threshold segmentation and morphological operations, the features of candidate nodules are enhanced by super-resolution convolutional neural network. Then several methods including speeded up robust features, gray-level co-occurrence matrix and gray-level invariant moments are used to extract various local and global features of candidate nodules, and bag-of-words model is utilized for dimensionality reduction and fusion. Finally, XGBoost-decision tree classification model is used to remove false positive nodules and complete the detection of lung nodules. The experiments on LIDC-IDRI data show that the model can achieve an accuracy of 97.87% and a recall rate of 97.92%. The proposed model can be used to assist doctors in the diagnosis of lung nodules and has certain clinical application value.