Deep Learning-Based Algorithm For Brain Tumor Classification And Detection

Brain tumor is one of the deadliest cancers in the world today.According to the clinical research of the World Health Organization,early detection of brain tumor patients and accurate judgment of tumor type and lesion degree can greatly improve the cure rate of brain tumors.However,due to the complexity of human brain organs,the wide variety of brain tumor and unpredictable metastasis,the classification of brain tumor is highly dependent on the professional knowledge and clinical experience of doctors.Although the accuracy of manual inspection is high,doctors will inevitably make misdiagnosis in the face of massive image data,and manual diagnosis takes a long time.At the same time,there is a shortage of doctors with relevant knowledge.Therefore,the research on automatic classification and detection of brain tumor has important application value.Currently,with the development of medical technology,medical imaging is playing an increasingly important role in the medical field.Since the human brain is an extremely complex and fragile tissue,MRI is commonly used for the classification and detection of brain tumors due to its non-invasive nature and lack of exposure to ionizing radiation.In recent years,various neural network models and algorithms have emerged in the field of medical imaging to assist doctors in making medical decisions.In this context,this thesis carries out research on the brain tumor classification detection algorithm based on deep learning,and completes the following work:(1)A classification method of brain tumor image based on multi model deep le arning algorithm is proposed.This method performs multi-classification on brain tum or MRI images by three multi-model neural network architectures,Mobile Net V2+Bi L STM,VGG19+CNN and Efficient Net+Bi GRU.The experimental results show that th e combination model improves the classification performance by 0.08～1.4% compared to the single model,and the average accuracy of the optimal model classification is improved by about 0.02%～3.9% compared to existing methods.(2)A new hybrid brain tumor classification method based on CBAM(Convoluti onal Block Attention Module)and Efficient Net neural network(IC+IEffx Net)with im proved channel attention mechanism is proposed.This method is divided into two st ages,with the first stage being feature extraction using a CBAM model based on an improved spatial attention mechanism.In the second stage,the Squeeze and Excepti on(SE)blocks in the Efficient Net architecture are replaced with Efficient Channel A ttention(ECA)blocks,and the combined feature outputs from the first stage are use d as inputs to the second stage.The experiment shows the four classification results of images of glioma patients,meningioma patients,pituitary tumor patients and norm al patients in the mixed brain tumor MRI dataset.The experiment results show that the average classification accuracy is improved by 0.6% or more than the existing m ethods.The experimental results demonstrate the validity of the method and provide a new reference for medical professionals to be able to accurately determine the type of brain tumour.