Rain Leo

Undergraduate degree.

ZhengZhou University

Student

NLST (Learn more about this study)

NLST-789

Apr 21, 2021

Lung nodule detection technology based on weak supervised learning

Aiming at the problem of detecting pulmonary nodules from CT images, this paper specifically studies a nodule detection framework deepem based on three-dimensional deep convolution neural network, which is used to detect pulmonary nodules by mining a large number of weak supervised tags in electronic medical records (EMR). In this paper, EM algorithm is used to train the whole model end-to-end, and deep 3D convolutional neural network is used to parameterize the traditional em. Firstly, the probability learning model of pulmonary nodule detection is established, and each nodule suggestion is regarded as a potential variable. Under the condition of image and weak label, it is inferred that the posterior probability of nodule suggestion is true, and the posterior probability is further fed back to the nodule detection module for training. In this paper, we directly build the environment and build the code on the windows operating system, through training three data sets - luna6 data set as supervised detection, NLST data set as weak supervised detection, Ali Tianchi data set as independent test set, in order to achieve more accurate detection results.