Automatic segmentation method based on PCANet depth learning model

Abstract

The invention discloses an image automatic segmentation method based on a PCANet depth learning model. The method comprises the following steps that (1) the images to be segmented are sequentially processed by preprocessing, super pixel clustering, the first PCANet network and the first SVM classifier to obtain the coarsely segmented images; the coarse segmented image is binary pre-segmented image. (2) the coarse segmented image obtained in the step (1) is sequentially processed by a second PCANet network and a second SVM classifier to obtain a refined segmented image, and the refined segmented image is a binary final segmented image, thereby completing automatic image segmentation. By controlling the whole data processing flow of the automatic segmentation method and the frame structure of the corresponding automatic segmentation system, the method of the invention combines the super pixel with the PCANet network to realize the automatic image segmentation, and can greatly improve thesegmentation accuracy and robustness of the molybdenum target mass.

Description

technical field [0001] The invention belongs to the field of image segmentation in image processing and analysis, and more specifically, relates to an automatic segmentation method based on a PCANet deep learning model (specifically, an automatic segmentation method based on a two-stage PCANet deep learning model for a mammary tumor mass), Correspondingly, an automatic segmentation system based on the PCANet deep learning model can also be obtained, which is especially suitable for the automatic segmentation of mammography masses. Background technique [0002] Medical imaging technology is an important part of modern medicine and has revolutionary significance for the diagnosis and treatment of diseases. Taking breast cancer as an example, breast cancer has always been one of the malignant diseases that threaten the physical and mental health of women all over the world. Therefore, it is very important to detect and screen abnormal breast tissue at an early stage. In terms ...

AI Technical Summary

Problems solved by technology

[0004] At present, the methods used in the field of image segmentation are generally divided into the following types: the first type is based on the gray threshold method, which divides the image into the foreground area and the background area according to the histogram of the target image, and then according to the extracted The circularity, area, standard deviation and other characteristics of the region are used to remove false positive regions. However, this method has fewer feature indicators and relies too much on the initial grayscale information. When the environment and the image grayscale information of different patients change greatly , the segmentation effect is often very poor, so it does not have strong robustness, and is currently mostly used in the preprocessing part of the image

The second type is the edge-based method, represented by level set segmentation, which calculates the gradient information in the image to construct the energy functional function, and establishes a certain relationship with the contraction and expansion strength of the curve. The final segmentation result is obtained by border fitting, but this method is often very sensitive to the mutation interference information in the image and it is difficult to detect the tumor area with blurred borders. In addition, the selection of the initialization area of ​​​​this method is very critical and often requires human intervention. To achieve the true sense of automatic segmentation

The third type is a clustering-based method, which first specifies multiple cluster centers, and then classifies all the points in the image into it within a certain range, recalculates the new cluster centers, and repeats iterations until a certain criterion is met. Optimum achieves the final result, but due to the great difference in the size and shape of breast masses, a single clustering method cannot achieve the ideal segmentation accuracy

The fourth category is the segmentation method based on deep learning. The most typical is to use the convolutional neural network to extract features from the input image and finally achieve the segmentation effect. However, the convolutional neural network often faces the problems of huge parameter adjustment work and redundant data volume. And too deep network has the defect of gradient disappearance, so it is not conducive to the realization of accurate image segmentation

Images