A Hash Retrieval Method for CT Images of Pulmonary Nodules Based on Medical Signs and Convolutional Neural Networks

Abstract

The invention discloses a pulmonary nodule CT image Hash retrieval method based on a medical sign and a convolutional neural network. The method comprises the following steps: firstly, according to the annotation of a specialist to nine sign values of a pulmonary nodule, constructing an accurate binary code of a training set; secondly, extracting the significant semantic feature of the pulmonary nodule by using the convolutional neural network and a principal component analysis method, and reversely solving a Hash function by combining an accurate Hash code of the training set; and finally, providing a retrieval method based on a self-adaption bit, and realizing the rapid checkout of a pulmonary nodule image to be inquired. The pulmonary nodule CT image Hash retrieval method based on the medical sign of the pulmonary nodule and the convolutional neural network is capable of effectively shortening the inconsistency between the image bottom feature and the advanced semantic feature, realizing the rapid checkout of the pulmonary nodule image, and providing the decision support of diagnosing pulmonary lesions for a doctor.

Description

technical field [0001] The invention relates to the fields of image recognition and machine learning, in particular to a method for encoding and retrieving images of pulmonary nodules based on a convolutional neural network. Background technique [0002] The retrieval of lung medical images based on content similarity plays an important role in the computer-aided diagnosis of lung cancer. In recent years, binary hashing has attracted extensive attention due to its advantages of small storage space and fast matching speed. However, traditional hashing methods are often based on the underlying features of images designed by hand, and then learn hash functions. There is a big difference between the features based on manual design and the advanced diagnostic semantics of pulmonary nodules described by physicians, which cannot well describe the sign information contained in pulmonary nodule images. In addition, there are a large number of images in the database that are equal to...

AI Technical Summary

Problems solved by technology

There is a big difference between the manually designed features and the advanced diagnostic semantics of pulmonary nodules described by doctors, which cannot well describe the sign information contained in pulmonary nodule images

In addition, there are a large number of images in the database that are equal to the Hamming distance of the query image, and the returned images cannot be sorted using the traditional Hamming distance

Images