Research On MRI Image-assisted Diagnosis Of Alzheimer’s Disease Based On Convolutional Neural Network

The early diagnosis and treatment of Alzheimer’s Disease(AD)has important practical significance for alleviating social pressure and maintaining the physical and mental health of the elderly.The development and progress of medical imaging technology,especially Magnetic Resonance Imaging(MRI)technology,enables the use of algorithm-based computer-aided diagnosis technology to improve the work efficiency and diagnostic accuracy of imaging physicians.As the early stage of AD,Mild Cognitive Impairment(MCI)has problems such as uncertain areas of brain atrophy and insignificant atrophy at the time of diagnosis,resulting in low diagnostic accuracy.In order to make full use of the effective information of MR images,many studies have manually extracted features and divide the Region of Interest(ROI).These human-subjective methods require high prior theoretical knowledge and are time-consuming and labor-intensive,and may also cause errors due to manual screening.In response to the above problems,this paper used a method that combines Convolutional Neural Network(CNN)and attention mechanisms to automatically extract important feature information of the lesion area,identify and classify the period and extent of the lesion,so as to effectively assist the physician in making accurate diagnosis of the disease.First,analyzed the attention mechanism,this selective attention mechanism is consistent with the characteristic that physician focus their attention on the target lesion and ignoring background information when observing and recognizing MR images,it’s a common feature enhancement strategy in the field of image classification and recognition.Then studied the attention modules with different functions,analyzed their working principles and implementation methods.In this paper,CBAM(Convolutional Block Attention Module),which combined spatial attention module and channel attention module,is used to construct two kinds of CNN based on attention mechanism.By calculating the feature information extracted from different dimensions,the weight of attention is generated to adjust the proportion of the extracted features and ignoring the secondary factors to accurately extracting the lesion features with stronger characterization ability.On this basis,in order to enable the network to obtain richer attention features,the residual attention module is formed by combining three-layer residual learning module with CBAM.Further deepening the network can avoid the problems of vanishing gradient and excessive parameters,so that the improved network can pay attention to and learn more discriminative lesion features in MR images of AD.In this paper,we used MRI data from the Alzheimer’s Disease Neuroimaging Initiative(ADNI)database to simulate the algorithm.The MRI data was preprocessed in a targeted manner before the experiment.The experimental results and comparison with other methods show that this study has achieved good classification results in the auxiliary diagnosis of AD and MCI,and also provides new ideas for the diagnosis of other brain diseases.