Identification Of Benign And Malignant Liver Tumors Based On Multiple Instance Learning

The liver is the largest organ in the human body.It has been a major threat to human life and health.CT imaging diagnosis has always been an effective means of clinical detection of liver tumors.At present,most of the process of imaging diagnosis is that doctors make good and malignant diagnosis by observing images by the naked eye and combining medical knowledge and experience.Therefore,the diagnosis results have certain subjectivity and have a great relationship with the doctor's practice experience.Besides,the amount of medical image data is large.It is a huge workload for doctors to identify and identify images in limited time.Therefore,in the medical academic field,there is an urgent need for the computer aided diagnosis and treatment technology(CAD)that can diagnose and identify the medical image image and give the auxiliary reference.Computer aided diagnosis and treatment of benign and malignant liver tumors based on CT images is one of the difficult problems in the international research field.Computer aided diagnosis and treatment of liver tumors is of great importance for the identification of benign and malignant liver tumors.At the present stage,the effective diagnosis of liver cancer was punctured,but this method will cause great harm.With the development of computer and medical image processing technology,the statistical features,texture features and shape features of CT images can be analyzed in depth,and the image classification method can be used to identify the liver tumor effectively.In this paper,texture features,statistical features and shape features are extracted from medical CT images based in image feature technology.The algorithm based on multi-instance learning(MIL)is used to identify the malignant tumor of the liver.It provides a necessary and effective aid for doctors to identify the malignant and malignant liver tumors.In this paper,a multi-instance packet construction method for medical images is proposed.The ability to recognize the whole image as a whole is realized under the multi-instance learning algorithm,which greatly reduces the workload of the doctor's manual identification of the tumor area,and reduces the subjectivity and empiricism of the doctor.The error of the classified sample label.Secondly,the traditional classic multi-instance learning algorithm Citation-KNN is improved based on the distance weighting in the package space,and DWCKNN algorithm is proposed.The experimental results show that the improved algorithm achieves high accuracy in the classification experiment of the same data samples.