Joint And Deep Regression Of Clinical Scores For Alzheimer’s Disease Using Longitudinal Data

Alzheimer’s Disease(AD)is a progressive neurodegenerative disease that often occurs in the elderly population.The disease is usually accompanied by the loss of cognitive abilities,including daily activities and decision-making abilities,as well as the decline of social life abilities such as mobility disability,aphasia,and agnosia.Once the patient is diagnosed with AD,there is no treatment currently available to cure it or to alter its progressive course.At present,the clinical diagnosis of AD is mainly based on the combination of neuroimaging(e.g.,Magnetic Resonance Imaging,MRI)and various clinical scores(e.g.,Alzheimer’s Disease Assessment Cognitive subscale).Then,the doctor estimates the severity of the patient’s disease is in the early stage or late state,and takes corresponding treatment for him.However,it is time-consuming for neurologists to obtain the clinical scores of patients,and for patients with poor compliance,it is more difficult to obtain the score.With the continuous application of neuroimaging technologies(e.g.,MRI)in the early intelligent diagnosis of AD,clinical scores prediction via neuroimaging data is greatly desirable since it is able to reveal the disease status adequately.Besides,multiple time points data can observe the changes of the patient’s brain map information at different time points,so that more complete development information of AD and key information at a certain time point during can be collected,which is more conducive to monitoring the disease development.In fact,most previous studies are focused on a single time point without considering relationship between neuroimaging data(e.g.,MRI)and clinical scores at multiple time points.Differing from these studies,we propose to build a joint and deep learning framework based on longitudinal multiple time points data to predict clinical scores.The main contents are as follows:First,we propose a deep and joint learning of longitudinal data for Alzheimer’s disease prediction method,which can explore the relationship between the characteristics of brain regions in MRI and clinical score.More precisely,the proposed framework consists of three parts.Firstly,the Correntropy and fused smoothness term are added into the traditional group LASSO method to achieve the feature selection,and obtain the most informative features.Then the selected features are fed into the Deep Polynomial Network(DPN)for deep feature learning.Finally,support vector regression utilized to predict the clinical score at future time point.Hence,the proposed framework can help the doctors to make a preliminary diagnosis of the patient’s condition.Second,we propose a novel joint deep learning regression prediction method,based on the deep learning method Independently Recurrent Neural Network(Ind RNN)to realize the clinical scores prediction of AD,to further improve the prediction accuracy of clinical scores.Specifically,a feature selection method combining group LASSO and Correntropy is used to reduce the high dimensionality of MRI features,and obtain the most informative features.Then,we explore the connection between different brain regions and the relation of the longitudinal time points data by the multiple layers of Ind RNN.By establishing the joint deep learning network model which including feature selection and multilayers Ind RNN,we can realize the prediction of clinical score at future time points,and predict the possible value of clinical score of patients,which can enhance access to early diagnosis and treatment.In conclusion,this paper conducts a study on longitudinal multiple time points data to predict the clinical score of patients by combining with deep learning network.Two methods were carried out respectively.With the basic completion of the first method,the second task is to optimize the last method,and improve the prediction accuracy of clinical score.Extensive experiments are carried out to prove the effectiveness of the proposed method on the public dataset.We predict longitudinal clinical scores at future time points by the proposed framework.Besides,the corresponding Mean Absolute Error(MAE)and Pearson correlation(R)between the ground truth and the predicted scores are calculated to estimate the experimental results.The Extensive experiments demonstrate the MAE of the proposed model in this paper is 3.12,3.84,2.83 and the Pearson correlation value is 0.83,0.81,0.86 for the prediction of the Alzheimer’s Disease Assessment Cognitive subscale at 18 months,24 months and 36 months.