A Breast Mass Segmentation Method Based on Cross Attention Mechanism

Abstract

The invention relates to an X-ray assisted diagnosis technology, and aims to provide a breast mass segmentation method based on a cross-attention mechanism. Including steps: making a data set; preprocessing with a cross-attention mechanism; constructing a deep convolutional neural network; adjusting the pre-training weight distribution of the image network dataset image network dataset; using the data preprocessing results to adjust the pre-training weight distribution, Train the deep convolutional neural network; use the deep convolutional neural network to infer the X-ray images to be detected. The present invention uses the cross-attention mechanism to quickly train the model, and the cross-attention mechanism enables the network to select tumor features from the two orientations of the MLO position and the CC position, and learn to adjust the weight value of the cross-attention. Compared with the traditional way of judging tumors only from CC or MLO positions, this invention can quickly and efficiently segment out tumors on mammograms, improve the detection rate and accuracy of tumors in mammograms, and has a high practical Application and promotion value.

Description

technical field [0001] The invention relates to a breast X-ray auxiliary diagnosis technology, in particular to a breast mass segmentation method based on a cross attention mechanism (CrossAttention). Background technique [0002] Breast cancer has developed into one of the most common cancers in women. Accurate identification of breast lumps is a necessary condition for the diagnosis of breast diseases. Due to the interference of high-density and iso-density glandular tissues in the breast, it is extremely difficult to identify the lumps in mammary X-rays with the naked eye. A lot of energy and time for doctors. Using artificial intelligence to learn a large number of tumor X-ray samples can read the images faster and locate the tumor location more accurately, laying a solid foundation for doctors to determine whether the tumor is benign or malignant. [0003] In clinical work, the bilateral breasts are generally examined by radiography, and usually two positions are requ...

AI Technical Summary

Problems solved by technology

In the diagnosis of breast X-ray, it is necessary to combine the MLO and CC positions to comprehensively judge whether there is a mass in the image. However, the existing technology can only judge the mass from one of the MLO or CC positions, which is very easy to cause missed detection and false detection. Clinically required accuracy

Images