Research On Class Prediction Models Of Parkinson’s Disease Through Multimodal Information Fusion And Machine Learning

Parkinson’s disease is a chronic neurodegenerative disease in the middle-aged and elderly,which leads to patients’motor and non-motor disabilities due to loss of dopaminergic neurons in the brain and affects patients’quality of daily life.With the rapid increase of aging population in China,the number of people with Parkinson’s disease has increased significantly,which has brought a heavy burden to the family and society.The timely,accurate and effective detection of Parkinson’s disease is not only a medical problem,but also a social and public health problem,which is essential for subsequent medical intervention and health management.This study aims to build multimodal data-driven models for Parkinson’s disease class prediction and lay a solid foundation for the development of intelligent decision support systems.（1）A cascade ensemble learning model for Parkinson’s disease class prediction based on handwritten dynamics exam was constructed.First,a cascade ensemble learning model fusing multiple basic classifiers was studied and the proposed model achieved satisfying class prediction accuracy on two public human activity recognition datasets.Furthermore,a novel cascade ensemble learning model fusing two Random Forest classifiers and two Extremely Randomized Trees classifiers in each layer was proposed to differentiate Parkinson’s disease patients from healthy individuals based on handwritten dynamics exam.To improve the classification performance,Principal Component Analysis technique was employed to reduce the data dimensionality.Experimental results showed that the proposed model achieved reasonable performance with 81.17%accuracy.It is inferred that the proposed model has certain significance for Parkinson’s disease identification.（2）A novel Parkinson’s disease class prediction model was proposed based on multimodal handwritten dynamics exams.First,six different Random Forest models were separately constructed to generate the corresponding class probability vectors which represent an individual’s class predictions on 6 different handwritten exams,and the final prediction result for an individual was obtained through voting strategy of all Random Forest models.The constructed model based on multiple handwritten exams has satisfying accuracy（89.4%）,specificity（93.7%）,sensitivity（84.5%）and F1-score（87.7%）.It is found that the class prediction performance based on spiral exam is optimal compared to the other handwritten exams.The class prediction model based on multimodal handwritten dynamics exams has better Parkinson’s disease identification performance and can be used as a potential and cost-effective approach for Parkinson’s disease screening.（3）The author developed early Parkinson’s disease class prediction models using Movement Disorder Society-Unified Parkinson’s Disease Rating Scale and Da TSCAN(¹²³I-Ioflupane)Single Photon Emission Computed Tomography.The concrete steps contain data collection,feature selection,classifier modeling,and early Parkinson’s disease class prediction.Total 61 statistically significant features between early Parkinson’s disease and healthy control group were considered as input to construct classification models.We observed that the constructed models achieved promising classification performances with accuracy,sensitivity,and Area Under Curve all≥98.32%and specificity≥96.48%.Meanwhile,3-class prediction models were proposed for the first time to differentiate early Parkinson’s disease from healthy control and early Scan Without Evidence of Dopamine Deficit using multimodal features composed of 59 scale features and 14 medical imaging features,with 95.31%accuracy obtained.It is found that clinical scale and Single Photon Emission Computed Tomography are both important for early Parkinson’s disease identification.The constructed models can be considered as effective methods for early detection and diagnosis of Parkinson’s disease,which are more applicable and valuable.The research is helpful for optimizing the allocation of health resources and improving the quality of life of Parkinson’s disease population.