Machine Learning Based Automated Classification,Retrieval And Prediction For Alzheimer's Disease

Alzheimer's disease(AD)is a neurodegenerative and non-curable disease.To better explore the underlying disease patterns,a study of modeling the disease procession for early diagnosis,retrieval and clinical scores prediction is necessary.(1)To better diagnosing AD or its early stage—mild cognitive impairment(MCI),we proposed a new method to represent the multi-modal correlations of regions of interst(ROIs)for AD diagnosis.Firstly,we applied canonical correlation analysis to discover the relationships of ROIs among different modalities and specifically with sparse least square regression loss functioin to select the discriminative features.Then we trained a classification model via support vector machine by using the selected features.We can draw a conclusion that the proposed method is effective to boost the early diagnosis of the disease.(2)Neuroimaging retrieval have become general in computer aided diagnosis.To handle it,an adaptive ensemble manifold learning(AEML)is proposed to retrieve similar brain condition via multi-source neuroimaging data.A novel objective function is devised via a manifold learning strategy to learn similarity by the geometrical constraints.The conplementary power of various source of data for discovering brain disease disorder is investigated via various weights.Additionally,a generalized norm is explicitly explored to control sparseness degree adaptively.Extensive experimental results demonstrate that our algorithm obtains quite appealing and superior performance.(3)Since there are missing data in different modality in Alzheimer's dataset,we propose a novel data complement and multi-time-series regression framework.We assume that matrix is low rank intrinsically,in that way a generalized low rank matrix approximation can recover the missing data.Then we use the recurrent neural network,which is good at processing time-series data,to train a multi-class classifier for diagnosing AD/MCI/NC and a regressor to predict the clinical scores.In summary,research objective of this paper is computer-aided diagnosis of Alzheimer's disease.Based on machine learning methods,classification,retrieval,and prediction are three main tasks conducted respectively in this research.The researches focus on several of the most critical issues in diagnosis process of AD,including data of heterogeneous source and heterogeneous structure and development of multiple time-points.Based on public databases,a large number of experiments results demonstrate the feasibility and.effectiveness of the proposed method.At the same time,the theories,models and algorithms proposed in this paper also have guiding and reference significance for the auxiliary diagnosis and application of other dementia types or neurological diseases.