DBT lump automatic segmentation method based on expansion depth convolutional neural network

Abstract

The invention discloses a DBT lump automatic segmentation method based on an expansion depth convolutional neural network, and the method comprises the steps: collecting an image, preprocessing the image, extracting image blocks, and automatically segmenting a DBT lump: employing an expansion depth convolutional neural network segmentation model to classify the central voxels of all image blocks extracted in the step 3, obtaining the probability distribution of each voxel corresponding to the two categories of the normal breast tissue area C0 and the breast lump area C1, wherein the category with the maximum probability serves as the category to which the voxel belongs, and therefore the volumes of the tissues corresponding to the normal breast tissue area C0 and the breast lump area C1 are obtained; and finally obtaining a final segmentation prediction result of the DBT lump and outputting the prediction result in the upper computer. According to the method, the expansion deep convolutional neural network is adopted, and the pooling filter is replaced by the expansion filter for structure optimization, so that the accuracy and robustness of automatic segmentation of the DBT lump can be improved.

Description

technical field [0001] The invention relates to the fields of digital medical image processing and analysis and computer-aided diagnosis, in particular to a DBT mass segmentation method based on an expanded deep convolutional neural network. Background technique [0002] Breast cancer has become a common malignancy among women worldwide and is the leading cause of cancer death in women. Mammography is the main method for clinical breast cancer detection at present, but it projects the three-dimensional breast entity to a two-dimensional plane. Due to the interference of overlapping breast tissue, the detection true positive rate and true negative rate are not high, which will lead to unnecessary These problems are more prominent in dense breasts. Digital breast tomosynthesis (DBT) is a new imaging mode developed based on the existing mammography technology, which can effectively reduce the misdiagnosis caused by overlapping tissues and improve the diagnostic ability of lesi...

AI Technical Summary

Problems solved by technology

However, these three types of automatic segmentation methods all need to artificially understand the image and then manually design and extract a large amount of specific feature information. There is no unified standard for the selection of these features, which largely depends on experience, and the extraction of features is computationally intensive and expensive. Time

Images