An Attention Model Based CNN For Alzheimer’s Disease Assisted Diagnosis

Alzheimer Disease(AD)is the most common disease among the elderly.With the aging of the world population,the incidence rate is increasing every year.Mild cognitive impairment(MCI)is the early manifestation of AD which can be divided into stable mild cognitive impairment(sMCI)and transitional mild cognitive impairment(cMCI).However,the symptoms of MCI are not obvious,and it is often mistaken for normal aging,which makes the patient miss the best period of treatment.Therefore,the diagnosis and early intervention of MCI patients and early AD patients are of great significance for delaying the development of AD.Medical imaging technology(MRI)has become a popular tool in the field of brain research,which can clearly reflect the internal structure of the brain and play an important role in the early diagnosis and prognosis of AD.In this paper,we use MRI dataset and deep learning network to obtain high-precision classification model.Based on the different dimensions of the data,we propose two different AD recognition methods,which are based on 3DMRI image data and 2DMRI slice data.In the method based on 3DMRI data,firstly,the dataset is further preprocessed,including skull removement and data enhancement.We proposed a 3D convolutional neural network,the 3D convolutional neural networks based on residual module and dense module,a 3D convolutional neural network based on densely connected attention mechanism.In this paper,the MRI data of normal control(NC)were used as the control group.The accuracy rates of AD vs.NC,NC vs.cMCI,and cMCI vs.sMCI reached97.15%,88.82%,and 78.79%,respectively.For 2DMRI data,we firstly sliced the 3DMRI along the coronal plane,and obtained80 2DMRI images each sample.Then,a dual-backbone composite network,a dual-backbone network with channel attention mechanism,and a dual-backbone network based on convolution kernel selection attention mechanism are designed for classifying 2D images.Features are extracted through the above network,and then put into an LSTM with a self-attention mechanism.Through the five-fold cross-validation method,the accuracy of the network proposed in this paper is 97.67%,74.86%,and 67.15% in AD vs.NC,NC vs.sMCI,and cMCI vs.sMCI respectively.In short,we designed multiple deep convolutional neural network models based on the attention mechanism to study the MRI data of Alzheimer’s disease,and explored thedifferent classification of images by different network models.According to the results,the3 D convolutional neural network based on densely connected attention mechanism and dual-backbone network based on convolution kernel selection attention mechanism in our proposed model have the best results.