改进的机器学习模型在肺结节良恶性分类中的研究

Improved Machine Learning Model in the Classification of Benign and Malignant Lung Nodules

肺癌发病率不断上升,借助影像手段快速进行肺结节的早期评估,对提高患者生存质量具有重要意义。针对这个问题,提出了一种新的肺结节良恶性分类模型,该模型首先采用过采样算法消除良性样本比例过高导致的结果偏移现象;然后提取每个结节的影像组学特征,并结合斯皮尔曼相关性变量剔除以及最小绝对收缩选择算子进行特征筛选,选取最优特征子集;最后采用余弦递减型惯性权重改进随机发生分布式延迟粒子群算法,以精准搜索全局最优参数,建立最佳分类模型。利用LIDC数据库上的608例训练集和68例测试集对模型进行训练和测试。模型在测试集上的AUC、准确率、精确率、召回率分别为0.93、0.941、0.917以及0.971。结果表明该模型能有效分类肺结节,有望在临床上进行肺结节的良恶性辅助诊断。

With the increasing incidence of lung cancer,rapid early assessment of lung nodules by means of imaging is of great significance to improve the quality of life of patients.To solve this problem,a new classification model of benign and malignant pulmonary nodules is proposed.Firstly,the model adopts the oversampling technique to eliminate the result deviation caused by the high proportion of benign samples.Then,the image omics features of each nodule are extracted,and the optimal feature subset is se⁃lected by using spearman correlation variable elimination and minimum absolute contraction selection operator.Finally,by using the inertia weight of decreasing cosine,the random generation distributed delayed particle swarm optimization algorithm is improved to search the global optimal parameters accurately and establish the best classification model.The model is trained and tested on 608 training sets and 68 test sets on LIDC database.The AUC,accuracy,precision and recall rates of the model on the test set are 0.93,0.941,0.917 and 0.971,respectively.The results show that this method can classify pulmonary nodules more effectively and is ex⁃pected to be used in clinical diagnosis of benign and malignant pulmonary nodules.