Research On Computer Tomography Images Of Pulmonary Nodules Based On Deep Hash Technology

The early diagnosis of pulmonary nodules can effectively assist doctors to distinguish the early stage of lung cancer,which is of great significance to prevent the occurrence of lung cancer.Clinical doctors mainly judge the symptoms of pulmonary nodules through CT images.However,in the face of the explosive growth of the number of lung CT images,doctors need to spend a lot of time to distinguish to meet the clinical needs.Due to doctors’ different cognition of the images,misdiagnosis and missed diagnosis may also exist in the specific discrimination process.Medical image retrieval technology can retrieve CT images most similar to the current image pathological features from a large number of confirmed cases,assist doctors in discrimination and diagnosis,reduce doctors’ workload,significantly improve the discrimination accuracy and reduce the possibility of misdiagnosis and missed diagnosis.It has become one of the key technologies to improve doctors’ diagnosis of lung cancer.In order to solve the above problems,further assist the doctor in distinguishing,this paper studies the retrieval method based on deep hash technology on the basis of following the medical characteristics of pulmonary nodule CT image.Deep hashing technology combines the low-cost and high-efficiency characteristics of image hashing algorithm with the advantages of deep learning method in feature extraction and expression,which can effectively improve the accuracy of medical image retrieval.Firstly,a supervised CT image retrieval method of pulmonary nodules based on image segmentation and depth hash is designed.Aiming at the problem that the current mainstream medical image retrieval algorithms lack the ability to represent the high-level semantic features and key local features of medical images,this method first uses the image segmentation network combined with attention mechanism to extract the high-level semantic features and key local features of images,and a coding network based on depth hash is used to encode the segmented results to realize the transformation from image to hash coding,finally,hamming distance is used to compare the similarity between the target image and the query image.Secondly,an unsupervised CT image retrieval method of pulmonary nodules based on generating confrontation network and depth hash is designed.In view of the scarcity and subjectivity of medical tags in practical application,this method first encodes the image by using deep hash network to realize the transformation from image to hash coding,then the generated hash code is trained and learned by using the generated countermeasure network,and the optimization strategy is designed to make the improved hash code have stronger representation ability,finally,aiming at the discrete value of hamming distance calculation,an improved weighted hamming distance is designed to realize fine-grained sorting.Experiments show that compared with other mainstream algorithms,the proposed supervised algorithm achieves better results in various indexes such as retrieval accuracy under different length hash coding,and the image segmentation processing in this method can promote the retrieval accuracy of other methods.The proposed unsupervised algorithm can achieve the retrieval accuracy of hash codes with different lengths without using label information,and the results are close to those of the supervised method.Aiming at the shortcomings of the current methods and the problems existing in practical application,this paper designs the corresponding improved optimization algorithm,which realizes the significant improvement of retrieval accuracy and other indicators,so as to more efficiently assist doctors in the discrimination and diagnosis of pulmonary nodules,reduce doctors’ workload,and play a positive role in the process of clinical medical diagnosis and computer-aided medical treatment..