CT lymph node detection system based on space-time circulation attention mechanism

Abstract

The invention discloses a CT lymph node detection system based on a space-time circulation attention mechanism, and particularly relates to the technical field of medical image analysis. According tothe invention, based on the deep convolutional neural network and the circular attention mechanism, the attention feature map adaptive to the focus size can be constructed in the slice direction and the spatial domain of the lymph node CT sequence. The method comprises the following steps: firstly, extracting high-level spatial features corresponding to a lymph node CT image by using a pre-trainedconvolutional network; secondly, in a spatial neighborhood, constructing a recurrent attention mechanism based on a Gaussian kernel function by taking a lymph node central slice as a reference; on the basis, implementing a time (slice direction) attention mechanism based on a Gaussian mixture model; in addition, constraining the predicted attention position according to prior information of position distribution of the lymph nodes in the CT slice sequence; and finally, enabling the recurrent neural network to integrate the high-level features extracted by the two attention methods for classification to obtain a lymph node detection result.

Description

technical field [0001] The invention relates to the technical field of medical image analysis, in particular to a CT lymph node detection system based on a spatiotemporal cycle attention mechanism. Background technique [0002] When applying the deep learning model to the field of medical image analysis, in view of the special imaging characteristics of medical images, the model is often affected by the different sizes, shapes, scales, imaging quality, background tissues and organs of the medical analysis target, and the analysis accuracy remains to be seen. improve. [0003] The visual attention mechanism can imitate the unique attention mechanism of the human eye to visual information, and can accurately locate the region of interest, and further overcome the influence of irrelevant information on the model. In view of this, the present invention applies the traditional natural image-oriented deep visual attention model to lymph node CT sequence images, and classifies the...

AI Technical Summary

Problems solved by technology

[0002] When applying the deep learning model to the field of medical image analysis, in view of the special imaging characteristics of medical images, the model is often affected by the different sizes, shapes, scales, imaging quality, background tissues and organs of the medical analysis target, and the analysis accuracy remains to be seen. improve

Images