基于深度学习的肺结节多分类方法

临床上肺结节的评估往往需要综合临床信息和影像特征进行判断,不同类型的结节的肺癌概率和判定标准也不尽相同,文章基于LIDC-IDRI数据集和额外的人工标注,提出了一种肺结节多分类的方法,利用多分类卷积神经网络,对预处理之后的CT数据进行肺结节的四分类,得到的分类结果更注重对临床医生的可理解的特征分类进行判断。实验表明,该方法取得了良好的效果,四种分类的准确性都在92%以上。该方法可以给医生提供一个可靠的结节分类效果,便于后续的肺结节评估。

Multiple classification methods for pulmonary nodules based on deep learning

Clinical assessment of lung nodules often require integrated clinical information and imaging characteristics of judgment,the different types of nodules of the lung cancer have different probability and decision criteria,based on the LIDC-IDRI data sets and additional manual annotation,this paper proposes a classification of pulmonary nodules more method,using convolution neural network classification,lung nodules on CT data after pretreatment of four classification,the classification results more attention to clinical doctors to understand the characteristics of the classification.Experimental results show that this method achieves good results,and the accuracy of the four classifications is more than 92%.This method can provide a reliable classification effect for doctors and facilitate the follow-up evaluation of pulmonary nodules.