Research On Medical Image Processing Methods Based On Tensor Neural Network And Ensemble Learning Prediction Model

With the development of imaging technology in recent decades,magnetic resonance image had been fully applied to various clinical application scenarios in hospitals.Magnetic resonance images in clinical diagnosis can provide unnecessary non-invasiveness the real-time organ status display to patients,which can provide the necessary information support for the early detection of the patient's condition and the choice of treatment plan.Brain magnetic resonance imaging can provide a view of brain activity status,so brain magnetic resonance imaging has gradually become the main data form of brain research.Magnetic resonance,images have image attributes such as high resolution and large number of image voxels,which greatly limits the application of traditional machine learning methods based on feature selection and classification in brain magnetic resonance image analysis.With the gradual increase in the incidence of brain tumor diseases,the clinical demand for predicting overall survival of brain tumor diseases is also becomina increasingly prominent.Survival analysis based on magnetic resonance imaging can not only provide necessary information support for the timely adjustment of patients'treatment plans,but also provide patients with reference psychological treatment expectation.In view of the problems faced by the above-mentioned medical image processing represented by magnetic resonance and some related clinical needs,this paper mainly studies the medical image processing methods based on tensor neural network and ensemble learning overall survival prediction model.The main innovations and contributions of this paper are mainly in the following two aspects:(1)A framework based on Tensor Neural Network for fMRI classification is proposed.The algorithm framework used tensor neural network to build a shallow and wide magnetic resonance image classification framework.It can not only extract beneficial features from whole brain functional magnetic resonance images to improve brain state analysis,but also greatly compress the neural network to solve the"dimension disaster"problem in magnetic resonance image analysis,thus which breaking the limitation on equipment requirements in the deep learning system currently applied to magnetic resonance analysis.We verified this algorithm on the CMU dataset.The experimental results showed that our proposed framework based on Tensor Neural Network for fMRI classification is superior to the traditional classification framework.(2)A framework based on ensemble learning for overall survival based on ensemble learning predictive model is proposed.We firstly extracted the multi-modal and multi-angle features of the structural magnetic resonance images in this algorithm framework,and then selected the univariate features through the Kaplan-Merier survival curve to select the beneficial features.Finally,the beneficial features of the screening were inputed into the ensemble learning predictive model to complete the overall survival prediction.We verified this algorithm on the Brats2018 dataset.The experimental results confirmed that the framework based on ensemble learning for overall survival based on ensemble learning predictive model proposed in this paper is superior to the traditional machine learning prediction model.