Medical Image Retrieval Method Based On Multimodal Feature Fusion

With the rapid development of society and the continuous popularization of modern medicine,medical digital imaging devices such as computed tomography(CT)and magnetic resonance imaging(MRI)have been widely applied in clinical work and auxiliary diagnosis.And a significant effect has been achieved.In clinical diagnosis,by finding a medical image similar to the patient's lesion image,the doctor can refer to previous diagnostic experience and make a diagnosis faster and more accurately.At the same time,patients can also have a better understanding of their own diseases,and cooperate with the treatment better.The most widely used image retrieval technology is content based image retrieval.By retrieving the visual content features of images,the similarity measurement between images can be transformed into the similarity measurement between image features.However,medical images have special characteristics: firstly,the properties of medical ontology is very complex,and it is often difficult to be found and described.Secondly,medical images have strong semantic information,and there is a great "semantic gap" between the content and the semantic features of the image.Finally,the retrieval precision for medical images is very high.Therefore,the general image retrieval method can not be used directly for medical images.Aiming at the particularity of medical images,in order to improve the effectiveness of medical image similarity retrieval,a medical image retrieval method based on multimodal feature fusion is proposed in this paper.Considering doctors' attention to the location of the lesion and the type of disease during the diagnosis of the CT image of the brain,the location of the lesion and the type of the lesion are taken as two modes in this section.The similarity between images can be obtained by weighting the similarity of images under different modes,and the weighted weights are adaptively obtained by information gain based on attribute selection.And the similarity of images in different modes can be learned by training the global and local features of the image.The experimental results show that the accuracy of medical image retrieval using multimodal feature fusion is better than the single feature retrieval method,and its efficiency is high.In addition,a medical image retrieval method based on Markov random field is proposed on the basis of multimodal feature fusion in this paper.By constructing Markov random field for medical images,this method can not only consider the similarity relationship between retrieved images and image database images,but also take into account the similarity relationship between image bases.For different image libraries,different Markov random field construction methods are used,and the construction process is based on multimodal feature fusion similarity.At the same time,an approximate belief propagation algorithm is used to implement the retrieval,and a reordering strategy is used to further optimize the retrieval effect.The experimental results show that the medical image retrieval method based on Markov random field can improve the accuracy of retrieval and have good efficiency.