Method for establishing lung nodule segmentation device based on Mask-RCNN neural network

Abstract

The invention discloses a method for establishing a lung nodule segmentation device based on a Mask-RCNN neural network. The method comprises: establishing a training sample: successively subjecting an acquired three-dimensional lung CT image to cutting, data enhancement, and hard negative sample mining to obtain a training sample set; establishing a lung nodule segmentation network that includesan input layer, a first largest pooled layer, a 64*64*64 convolution layer, a second largest pooled layer, a 32*32*32 dense block Layer, a third largest pooled layer, and a 16*16*16 dense block layerwhich are successively connected, up-sampling the feature map output by the 16*16*16 dense block layer to be fused with the feature map output by the 32*32*32 dense block layer,and inputting the feature map subjected to the feature fusion into a RPN network through a POL pooling layer; training the lung nodule segmentation network: training the lung nodule segmentation network by the training sample to obtain the lung nodule segmentation device.

Description

technical field [0001] The invention belongs to the field of image processing, and in particular relates to a method for establishing a lung nodule segmentation device based on a Mask-RCNN neural network. Background technique [0002] There are many existing methods for detecting pulmonary nodules in lung CT images using deep learning algorithms, but the detection accuracy is not high. The main reasons for the low accuracy are: [0003] (1) The recall rate in the detection stage is lower than that of some special types of pulmonary nodules, resulting in missed detection and low detection accuracy. [0004] (2) The size of pulmonary nodules is uneven, and smaller pulmonary nodules are easily overlooked. [0005] Based on the above two reasons, the typicality and representativeness of pulmonary nodules detected and segmented by deep learning algorithms are insufficient. [0006] Therefore, improving the accuracy of pulmonary nodule detection and training the network to segm...

AI Technical Summary

Problems solved by technology

[0003] (1) The recall rate in the detection stage is lower than that of some special types of pulmonary nodules, resulting in missed detection and low detection accuracy

[0004] (2) The size of pulmonary nodules is uneven, and smaller pulmonary nodules are easily overlooked

[0005] Based on the above two reasons, the typicality and representativeness of pulmonary nodules detected and segmented by deep learning algorithms are insufficient.

Images