A three-dimensional pulmonary nodule recognition method based on convolution neural network

Abstract

The invention relates to the field of medical image analysis and the field of depth learning, which is a three-dimensional lung nodule identification method based on a convolution neural network. Themethod comprises the following steps: a lung three-dimensional CT image data set is pretreated, and the pretreated CT image data set is divided into a training data set and a test data set; A neural network model combining DenseNet and SENet is established, and its super parameters are set. The training data set is imported into the neural network model, and the stochastic gradient descent algorithm and the learning rate decrease gradually are used to train the neural network model. After the model converges sufficiently, the model structure and weight parameters are saved and deduced, and thetrained neural network model is obtained. Using neural network model to test each group of three-dimensional CT images in the test data set, the recognition results of lung nodules were obtained. This method can be used to analyze whether there are pulmonary nodules and their specific locations in 3D CT images, which can solve the problem of low accuracy of pulmonary nodules recognition caused bygradient disappearance, gradient explosion or degradation of deep convolution neural network.

Description

technical field [0001] The invention relates to the fields of medical image analysis and deep learning, in particular to a three-dimensional pulmonary nodule recognition method based on a convolutional neural network. Background technique [0002] Lung cancer is one of the most common cancers with the highest death rate. Early diagnosis and treatment of lung cancer can greatly improve the five-year survival rate of patients. Pulmonary nodules are the manifestation of lung cancer in chest computed tomography (CT) images, so identifying pulmonary nodules from chest CT images is an effective means of detecting lung cancer. However, since a set of lung CT images has hundreds of images and the size of pulmonary nodules is small, this will undoubtedly increase the workload of doctors, thereby reducing the work efficiency of doctors, and may also cause misdiagnosis and missed diagnosis. [0003] Due to the rapid development of deep learning and the emergence of more and more CT da...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the deficiencies of the prior art, the present invention provides a three-dimensional pulmonary nodule recognition method based on a convolutional neural network, which can analyze whether the three-dimensional CT image contains pulmonary nodules and the specific location of the pulmonary nodules, and mainly solves the problem of In the existing technology, due to the problem of gradient disappearance, gradient explosion or degradation that may occur in the deep convolutional neural network, the accuracy of pulmonary nodule recognition is low.

Images