Three-dimensional neural network Pulmonary nodule detection method and system based on slice perception

Abstract

The invention discloses a three-dimensional neural network pulmonary nodule detection method and system based on slice perception. The method comprises the following steps: acquiring a three-dimensional CT image; constructing a three-dimensional neural network according to the grouped slice non-local module, the three-dimensional region candidate network and the false positive reduction module, and obtaining a pulmonary nodule detection result of the three-dimensional CT image according to the three-dimensional neural network; extracting image features of a three-dimensional CT image by adopting the grouped slice non-local module; obtaining pulmonary nodule candidate targets according to image features by adopting the three-dimensional region candidate network; and using the false positivereduction module to obtain multi-scale features of the pulmonary nodule candidate targets, and fusing the multi-scale features to obtain a pulmonary nodule detection result. The invention includes learning a non-local dependency relationship among elements in the grouped slices through a grouped slice non-local module so as to better learn discriminative features; multi-scale features are extracted and fused through a false positive reduction module, so that pulmonary nodules which are wrongly recognized are reduced, and the pulmonary nodule detection effect is remarkably improved.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a method and system for detecting pulmonary nodules based on a slice-aware three-dimensional neural network. Background technique [0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art. [0003] In recent years, with the continuous development of computer technology and the wide range of needs in daily life, image processing technology has made vigorous progress. Target detection technology in image processing is used to detect specific target categories in images, and has many applications in natural images and medical images. Lung cancer has become one of the leading causes of cancer death worldwide. Pulmonary nodules are lesions in the lungs that have a high possibility of developing into malignant tumors. Early diagnosis of pulmonary nodules and timely treatme...

AI Technical Summary

Problems solved by technology

This type of method ignores the deep connection between different slices in 3D CT data, and cannot cope well with pulmonary nodules with many different shapes.

In addition, such methods do not consider the false positive problem that is very important in medical image processing, and there is no module to reduce false positive pulmonary nodules, so the effect is relatively poor when detecting small pulmonary nodules

Images