Lung CT Image Similarity Retrieval Method Based On Weakly Supervised Similarity Evaluation Network

With the rapid development of medical image technology and the advent of the era of big data,as one of the important types of medical images,the number of high-resolution lung CT images has shown an explosive growth trend.Finding images similar to the CT images submitted by users from the massive lung CT image database can effectively help doctors achieve auxiliary diagnosis.However,the similarity retrieval of lung CT images has higher accuracy requirements than ordinary content-based similarity retrieval,and has similar requirements for the similarity of external shape,internal blood vessels and lesion positions.In traditional supervised deep learning networks,the learning of the network relies on labels.However,the labeling of medical images requires professional medical personnel to annotate each image in a time-consuming and laborious manner,and the cost is too high.This paper proposes a deep learning network for similarity analysis of lung CT images based on weak supervision—WSSENet(Weakly Supervised Similarity Evaluation Network).Experiments show that the learning network has achieved satisfactory results in calculating the similarity measure and retrieval of lung CT images.This subject mainly completes the following works:1)Research on related algorithms.The development background of medical image retrieval algorithms at home and abroad is studied,and the related algorithms of medical image retrieval systems are studied.2)Propose a medical image retrieval algorithm based on deep learning.This paper proposes a layered learning deep learning network WSSENet,which evaluates the similarity of lung CT images in the outline of the lung lobe and the internal details of the lung lobe,thereby achieving high-precision similarity evaluation.3)According to the characteristics of lung CT images,combined with the variant of spatial deformation network,an automatic labeling algorithm is proposed.This label is a coarse-grained label,which can automatically generate the training set and test set for training the deep learning network.4)By combining the i Distance algorithm,the efficient top-k retrieval of medical images under big data is realized.