Research On MRI Image Classification Algorithms And User System For Brain Tumors

Brain tumor is one of the culprits of health hazards,if the types of brain tumors are not identified in time and targeted treatment is carried out,the survival rate of patients will be reduced.Therefore,it is particularly important to identify the types of brain tumors as the premise of their treatment.The appearance of magnetic resonance imaging(MRI)technology has played a good auxiliary role in distinguishing the types of brain tumors.However,doctors’ diagnosis based on MRI images of brain tumors depends on clinical experience,and there are some problems such as misdiagnosis,missed diagnosis and low efficiency.At present,the identification of brain tumor types based on deep learning has been widely studied as a frontier topic.Therefore,on the basis of full investigation,this thesis studies the deep learning algorithm to improve the classification accuracy of brain tumors.In order to improve the internal structure of the algorithm model,more effective features are obtained from brain tumors MRI images to improve the accuracy of identifying brain tumor types.Based on the Res Net50 algorithm model,this thesis uses the ECA attention mechanism module to enhance the algorithm model’s attention to brain tumors images and channel information.For the convolution operation,the characteristics of features are extracted by fusing spatial information and channel information.The CBAM attention mechanism module is added to realize the focusing of the spatial information and channel information of the brain tumors images before the convolution operation.The introduction of multi-scale input allows the algorithm model to learn the local features of brain tumors at different sizes.Finally,according to the improvement strategy of multi-scale input combined with dual attention mechanism,three algorithm models of ECA-Res Net,CE-Res Net and MCE-Res Net are obtained,and the effectiveness of the improved algorithm is verified by means of successive verification and simulation analysis.In order to avoid repeated modification of the algorithm model and improve the accuracy of identifying brain tumor types,this thesis adopts an improved strategy of transfer learning combined with multi-scale input,and conducts simulation verification on the basis of the Res Net50 algorithm model and the Inception V3 algorithm model.At the same time,in view of the limitation of a single algorithm model for misjudgment of brain tumor types,the improved Res Net50 algorithm model and the Inception V3 algorithm model are weighted and fused on the output layer to alleviate the limitations of a single algorithm model.From the simulation results of different improved algorithm models mentioned above,it can be seen that the improvement of the internal structure of the algorithm model has increased the accuracy rate of brain tumors identification from 92.25% to95.03%,an increase of 2.78 percentage points.Combining transfer learning and multi-scale input Res Net50 algorithm model and Inception V3 algorithm model,the accuracy of the single improved algorithm model increased from 96.24% and 96.57%to 97.06%,respectively.This thesis also preliminarily designs a validating user system based on deep learning and can predict the type of brain tumors.At the same time,the system is validated based on the data set in this thesis,which lays a foundation for the further development of the application system.