Multimodal Magnetic Resonance Imaging Based Research On Parkinson’s Disease Brain Networks

Parkinson’s Disease（PD）is a neurodegenerative disorder characterized by extensive structural abnormalities in cortical and subcortical brain regions.The main symptoms of Parkinson’s disease are bradykinesia,decreased amplitude or speed of movement（or progressive retardation or stagnation）,and at least one of the main motor characteristics is rigidity or quiescent tremor,which will not only seriously affect the patients’ daily life,but also bring a heavy and serious mental and economic burden to the family and society.However,the pathogenesis of Parkinson’s disease is still unclear.Early diagnosis and curative effect prediction of Parkinson’s disease has become an urgent scientific research problem.Therefore,it is of great guiding significance to explore the neural and pathological mechanisms of Parkinson’s disease and to find biological markers and features that can identify the pathogenesis of Parkinson’s disease in clinical diagnosis and intervention.Studies have shown that cerebral morphological structure changes caused by PD causes brain function on the gray and white matter abnormalities,However,the functional network changes caused by white matter dysfunction and their association with Parkinson’s symptoms remain unclear,the effectiveness of these brain changes in predicting clinical surgical and therapeutic improvements deserved further investigation.In recent years,with the development of f MRI,the structure and function of neuroanatomical regions of the brain can be observed in a safe and non-invasive way.In this paper,with the help of MRI technology and related research methods of machine learning and deep learning,firstly takes the white matter of PD as the breakthrough point,from the perspective of white matter functional network to characterize the abnormal interaction pattern and discusse potential pathophysiological mechanisms of PD.Furthercombining multimodal data,the structural modal data of white matter and gray matter and behavioralscale were used to identify and explore the characteristics related to the improvement of symptoms after DBS,trying to explore white and gray matter multimodal features that can automatically identify PD imaging markers and have good predictive ability for surgical outcomes.The main research contents are as follows:1.Aiming at the abnormal interaction between white matter functional networks in PD,a total of 12 white matter functional networks were identified by clustering approach,and they were further divided into deep,middle and superficial layers.Using coefficient Granger causal analysis method,we found that the excitatory/inhibitory influence between the three-layer white matter network in PD was disrupted compared with HC,especially the extensive aberrant interactions between the deep network associated with the basal ganglia and other networks.Furthermore,interaction changes involving deep networks were negatively correlated with UPDRSIII.This study demonstrates that informational associations between white matter functional networks are disrupted in PD,which may contribute to understanding the neuropathology of PD and its progression throughout the nervous system from the perspective of white matter function.2.Aiming at the problem of intelligent prediction of postoperative improvement effect after DBS in PD,the 3D-Convolution Neural Network and multimodal fusion method were perform binary classification training on multi-center large-sample Parkinson’s data to identify and accurately classify PD.The features were transferred to the preoperative data of patients with DBS,and combined with the preoperative scale to predict the improvement of the surgical efficacy,the predicted results and the actual results were calculated and analyzed using the Pearson correlation coefficient,mean square error,R square and other indicators,showing that better predictive ability.The research also combines interpretability methods such as Gradient-Class Activation Mapping（Grad-CAM）to achieve accurate localization of key features.It was found that the features that can classify and predict postoperative efficacy are mainly distributed in the basal ganglia,motor cortex,deep frontal white matter,occipital lobe,temporal lobe,cerebellum and other regions related to the damaged neural circuits in PD.This study shows that multimodal brain feature patterns can provide complementary information for exploring the mechanism of Parkinson’s pathology,and provide ideas for establishing a new multi-sample disease diagnosis model,and achieve accurate prediction of postoperative improvement from preoperative data.