Intelligent Diagnosis Of Alzheimer’s Disease Based On Multi-modal Data Learning

Alzheimer’s disease(AD)is one of the leading causes of death in the elderly.Its early stage is mild cognitive impairment(MCI),which is mainly manifested in memory loss,judgment ability decline,and so on.In the stage of AD,the patients’ memory is seriously damaged,their emotions become irritable,and even they can’t take care of themselves in daily life.The disease has serious harm,which not only causes great pain to patients but also brings a huge burden to family and social medical treatment.At present,AD can’t be cured,so the prevention and early intervention of the disease have become the main means and goals to overcome the disease.Therefore,if we can identify MCI patients in the early stage,and effectively identify whether MCI patients will further convert to the AD stage,it will help patients receive treatment as soon as possible and improve the effectiveness of diagnosis and treatment.For the early computer-aided diagnosis(CAD)research of the disease,single-mode data are generally used for analysis.However,AD is not easy to detect,and single-modal data can not provide accurate lesion information.And each modal data has its advantages and disadvantages.Using only one modal of data will ignore the complementary information between modalities.Therefore,the use of single-modal data has a certain effect on the classification of the AD stage,but the recognition accuracy of the early stage is not high.With the development of medical technology and computer technology,the auxiliary diagnosis based on multi-modal data has attracted more and more attention,and it has also shown a better classification and recognition effect than single-modal data.To make full use of multi-modal data and improve the recognition effect of AD stage and MCI stage,two intelligent diagnosis methods of Alzheimer’s disease based on multi-modal data learning are proposed in this paper.The two tasks are as follows:(1)In the first work,we propose a classification method of Alzheimer’s disease based on multi-modal hypergraph convolutional(HGCN).Firstly,each subject is regarded as a node,and the KNN strategy is used to establish a hypergraph on each mode of the subject to construct the multi-modal features.Then HGCN is used to extract the multi-modal high-order depth features with small dimensions and high recognition.Finally,a multi-view fuzzy classifier is used to comprehensively classify the multi-modal depth features extracted upstream.Experimental research shows that the classification results of this method can be comparable with the existing methods in the classification task of the AD stage.In the classification task of the MCI stage,the results are significantly improved.(2)In the second work,a classification method of Alzheimer’s disease considers both individual and fusion features.Based on the first work,the fusion feature is added as a new view.The depth features extracted by the HGCN method in previous work are regarded as the individual features,and the original features fused by the low-rank multi-modal fusion(LMF)method are regarded as the fusion features.Finally,the individual features and fusion features are comprehensively used in the multi-vies fuzzy classifier for training and comprehensive decision-making.This method not only retains the independent information of each mode but also takes into account the relevant information between different modalities.Experiments show that after adding fusion features,the classification results are improved in the classification tasks of the AD stage and MCI stage.The features fused using LMF can classify better than using the simple concatenation method.In general,the two methods based on multi-modal learning proposed in this paper for the diagnosis of Alzheimer’s disease show certain advantages over the existing methods and have better potential application value.At the same time,the method proposed in this paper can also provide an important reference for follow-up research.