Research On Bone Tumors Diagnosis Based On Machine Learning

Bone tumor is a kind of tumor that occurs in bone system,and the death rate of malignant bone tumor is very high.If bone tumor can not be treated in time,patients often have the risk of amputation or even cancer cell proliferation.Early detection and timely treatment is the key to reduce the death rate of bone tumor.However,due to the low incidence,it is very difficult to obtain a large number of labeled samples of bone tumor images.This paper studies how to use machine learning method to realize the diagnosis of bone tumors on medical images of small samples.This paper first introduces the basic principles of machine learning,including the basic support vector machine and neural network algorithm,as well as migration learning and active learning algorithm.To solve the problem of high cost and difficult to get labeled samples,this paper proposes a model training algorithm based on the distribution of sample information.The purpose of this algorithm is to use as few labeled samples as possible to realize the diagnosis model of bone tumor.By calculating the amount of information of the samples,the method determines the value of the samples,marks the valuable samples,and sends them into the model to fine tune the model parameters.Through repeated iterations,the model with high accuracy is finally obtained.Then,aiming at the problem of small sample size and low diagnostic accuracy of bone tumor,an algorithm based on multi feature fusion is proposed.The purpose of this algorithm is to deeply mine the features of limited samples and improve the accuracy of bone tumor diagnosis.The algorithm extracts the surface features and fuzzy features of the samples.The surface features reflect the bone contour,tumor contour and bone density on the bone tumor image.The fuzzy features are extracted by depth neural network.The fusion features of the two samples are obtained by splicing,and the fusion features are used for multi-level classification diagnosis of bone tumors.According to the experimental results,compared with randomly selected samples for training,the model training algorithm based on the distribution of sample information can reduce the sample tagging by more than 50%.The accuracy of bone tumor diagnosis algorithm based on multi feature fusion is 99%,which is higher than the traditional method.Detection of benign and malignant can also reach more than 85%,higher than the use of unipolar classification.The two algorithms proposed in this paper can reduce the number of labeled bone tumor samples,improve the accuracy of bone tumor diagnosis,it can solve the problem of difficulty in labeling bone tumor samples and low detection accuracy to a certain extent.