Joint Regression And Classification Via Relational Regularization For Parkinson's Disease Diagnosis

Parkinson's disease(PD)is a progressive neurodegenerative disease that is common in elderly people.PD can be mainly identified with the appearing of motor symptoms and non-motor symptoms.To slow this disease deterioration,accurate PD diagnosis is an effective way,which alleviates mental and physical sufferings by the timely clinical intervention.So this paper proposes a joint classification and regression framework for PD diagnosis based on relational regularization,which is subdivided into two research points:First,this paper proposes a joint classification and regression framework for PD diagnosis via baseline multi-modal data.Specifically,this paper devises a unified multi-task feature selection model to explore multiple relationships among features,samples,and clinical scores for selecting the most informative features.Further,we perform the classification of PD as well as regress four clinical variables from the multi-modal data.Second,this paper proposes a new unsupervised feature selection method via joint embedding learning and sparse regression using longitudinal multi-modal neuroimaging data.Specifically,the proposed method performs feature selection and local structure learning,simultaneously,to adaptively determine the similarity matrix.Meanwhile,this paper constrains the similarity matrix to make its connected component equal to the number of classification for gaining the most accurate information of the neuroimaging data structure.The baseline data is utilized to establish the feature selection model to select the most discriminative features.Next,this paper exploits baseline data to train four regression models for the clinical scores prediction and a classification model for the classification of PD in the future time point.Extensive experiments are conducted to demonstrate the effectiveness of the proposed method on the public dataset.In our study,two future time points,both 12 months and 24 months data are used to evaluate the classification performance of the proposed method.Meanwhile,this paper considers four scores prediction and a label classification using a 10-fold cross-validation method to validate the performance of the proposed method based on baseline data.The experimental results demonstrate that the proposed method can enhance the performance in clinical scores prediction and class label identification in the longitudinal data and outperforms the state-of-art methods as well.