Method for detecting X-ray mammary gland lesion image based on feature pyramid network under transfer learning

Abstract

The invention provides a method for detecting an X-ray mammary gland lesion image based on feature pyramid network under transfer learning. The method comprises the steps: 1, establishing a source domain data set and a target domain data set; 2, establishing a deformable convolution residual network layer by a deformable convolution and extended residual network module; 3, establishing a multi-scale feature extraction sub-network based on a feature pyramid structure through a feature map up-sampling and feature fusion method in combination with a deformable convolution residual network layer;4, establishing a deformable pooling sub-network sensitive to the focus position; 5, establishing a post-processing network layer to optimize a prediction result and a loss function; and 6, migratingthe training model to a small sample molybdenum target X-ray mammary gland focus detection task so as to improve the detection precision of the network model on the focus in the small sample image. According to the method, a transfer learning strategy is combined to realize focus image processing in a small sample medical image.

Description

technical field [0001] The invention relates to the technical fields of medical image processing, deep learning and artificial intelligence, and in particular to a method for detecting X-ray breast lesion images by a transfer learning feature pyramid network. Background technique [0002] With the rapid development of medical digital imaging technology, medical image analysis has entered the era of medical big data. Lesion detection in medical image analysis is one of the cross-research topics between auxiliary diagnosis and computer vision. Traditional CAD technology uses image edges, textures, and statistical features related to signal strength, HOG, Haar, and SIFT. Realize some simple lesion detection tasks in the image, but the lesion area in the image shows the characteristics of various shapes and scales, which leads to the low detection accuracy and poor generalization of the traditional algorithm model, so most of the image analysis work still needs to be done. It i...

AI Technical Summary

Problems solved by technology

But there are some problems: ①This invention is a pixel-level semantic segmentation model, but the final segmentation map comes from the 8-fold upsampling of the semantic segmentation feature map. This method has a huge error and the segmentation outline is rough; ②This invention uses classical cross-entropy Loss is used to construct the loss function, and the convergence of confrontational network training is constrained by adding the Q regular term method; ③Expanded convolution is empty convolution, and 3 scaling factors are given in this invention; ④The feature extraction structure in the network takes the most time, and the Three parallel branches are used in the invention, and the calculation redundancy of the overall system is large

But there are some problems: ①The invention is based on a large number of existing lesion data for learning, and there is no explanation for the small sample problem; ②Random sampling is performed on the entire image under the two-dimensional Gaussian probability constraint to obtain the RoI area, and the obtained The proportion of accurate target areas is small, and the calculation redundancy is large; and the parameters are given according to prior knowledge, which cannot guarantee the optimization of parameters and affect the accuracy of the model; An end-to-end learning method; ④The invention uses SVM to establish a binary classification discriminator, and establishes a binary classification loss to guide network classification, lacks feedback adjustment for position regression, and cannot perform fine-grained classification and identification of lesions;

But there are some problems: ① The baseline of the invention uses SE-MobleNet; ② The invention derives the c2-c5 network layer output features from the baseline, generates p2-p6 feature maps respectively, and performs conventional delay after all the features are fused. This is a conventional pyramid model;

But there are some problems: ①The loss function uses cross entropy, but only the standard function is given, and the specific mathematical model and improvement of the loss function are not involved; ②The convolutional layer of the neural network is a conventional convolution module, and there is no strict Consider the impact of residuals on the network;

But there are some problems: ①The structure design of the invention Baseline uses the conventional residual module and the classic U-Net model; ②The use of dilated convolution in the network is an existing method, but the specific implementation method of its expansion is not clearly explained, different The size of the dilated convolution ratio is also an explanation; ③The introduction of the attention mechanism should add innovation to the network, but this part is hardly detailed; ④The invention does not give the "cross entropy loss function" practical application mathematical model and Related parameters;

[0015] To sum up, none of the above existing technologies can transfer their models to the detection task of small-sample X-ray breast lesions, so as to improve the detection accuracy of the network model for small-sample lesions

Images