Research On CT Image Retrieval Of Pulmonary Nodules Based On Deep Hash

The effective diagnosis of pulmonary nodules is of great significance for the physician to reasonably and accurately determine the type of lung cancer.In the face of explosive CT image data of the lungs,a large number of experienced radiologists are required to diagnose,that can meet the clinical needs,which will inevitably lead to misdiagnosis and missed diagnosis.The emergence of computer aided diagnosis technology can help doctors’ diagnosis and reduce the workload of doctors to a certain extent.Image retrieval technology with the rise of computer aided diagnosis technology,especially the technique of image retrieval using image hashing method,which can be well applied to medical image retrieval because of its small storage space and fast retrieval speed.In the process of retrieving similar nodule pulmonary nodules,the image hashing method combined with the deep learning method is widely used because of the traditional image retrieval technology based on feature extraction method fails to fully express all the medical information of the lung image.The image hashing method combined with the deep learning method is used to extract more comprehensive image features,such as high-level semantic features that can characterize key information of images.This paper learns from the advantages of the deep learning method,based on the image hashing technology,makes a reasonable improvement on the network framework used by the specific method.Then a deep hash-based CT image retrieval method for lung nodules was proposed.By designing an effective retrieval method,the data of a large number of unlabeled lung nodules which are not marked by the physician were fully utilized.At the same time,an effective algorithm was designed to study the key medical signs information of lung nodules that characterize the type of pulmonary nodules.The specific research contents of this paper are as follows:(1)Accurate retrieval of similar lung nodule images is of great significance for quickly identifying the benign and malignant pulmonary nodules while reducing the workload of physicians.In view of this,a pseudo-label based pulmonary nodule CT image retrieval method was proposed.The method first constructs a deep learning framework to learn the characteristics of the labeled lung data;The deep learning framework is then used to learn the unlabeled lung data features based on the pseudo-label hashing method to perform tagging of unmarked lung data within the network framework using pseudo-labeling techniques,while generating corresponding Hash code;Then design an efficient loss function to guide the entire network for efficient training;finally,use adaptive weighted similarity calculation method to retrieve similar lung nodule images.The experimental results show that the retrieval method proposed in this paper greatly improves the retrieval performance,and the retrieval effect is better than the traditional retrieval method and the traditional Hamming distance calculation,and achieves high accuracy and retrieval accuracy.(2)To solve the problem that the image features of lung nodules are not wellexpressed and the retrieval effect is not good in the CT image retrieval of pulmonary nodules for the traditional hashing methods in the presence of medical signs,a hashing method for improved the distance calculation to retrieve similar lung nodules is proposed.Firstly,the depth learning method is used to extract the depth features of the lung image and to classify the lung image,and the depth features of the lung image are mapped to the corresponding hash codes;Then,an image feature association representation is performed on each of the lung image in the training lung image class and the query image with the same probability value distribution.The improved distance calculation method proposed in this paper is then used to calculate the similarity between the candidate images;Finally,the most similar lung nodule image can be quickly returned.The experimental results show that the proposed method can quickly retrieve images with similar features of lung nodules,and achieve efficient retrieval of similar lung nodules CT images with high retrieval accuracy.