The Application Of Multi-modal Functional MRI Technology In The Diagnosis And Classification Of Clinical Phenotypes Of Parkinson’s Disease

BackgroundIdiopathic Parkinson’s disease (PD) is the second most common neurodegenerative disease, which is characterized by varying motor and non-motor deficits. Though limbs are heavily affected, PD is viewed to arise from the central neural system. As depicted in a classic pathophysiological model of PD, the progressive depletion of dopaminergic neurons in the substantia nigra disturbs the function of the basal ganglia motor circuit, which ultimately causes multiple motor symptoms; however, given the heterogeneity of PD symptoms and even some manifestations showing contrary effects, the long-standing PD pathology model has been instantly challenged.Among the clinical manifestations of PD, postural instability and gait difficulty (PIGD) and resting tremor are most common motor signs. PIGD symptoms are mainly present in patients with the PIGD phenotype, which are associated with a worse prognosis, less success with levodopa replacement therapy and a higher risk of non-motor complications. The PIGD phenotype is thus taken as a refractory subpopulation of patients with PD. Previous neuroimaging studies have pointed out that more severe white matter impairments were present in the PIGD phenotype, which may account for the pathogenesis of PD. In contrast to the PIGD symptoms, PD patients with resting tremor are considered to have a relatively benign disease course. For resting tremor, functional abnormality of the cerebellar-thalamo-cortical circuit has been thought as the underlying neural mechanism. Given the heterogeneity of clinical manifestations and pathophysiological mechanisms existing in PD, it is increasingly important to perform differential diagnoses and thus to provide different treatments for patients with PD.The multivariate pattern analysis based on multi-modal MR features has been widely used in studies of translational medicine according to its excellent classification performance and the capability of processing high-dimensional datasets. Our prior study has proposed an automatic classification with a high discriminative power to differentiate patients with PD from healthy people. In clinical practice, the meaning of constructing a classifier is to facilitate the differential diagnosis and thus the effective treatment protocols for distinct subtypes of patients with PD.MethodsStudy 1:PD patients with the PIGD and non-PIGD subtypes were recruited. All the patients underwent diffusion tensor MRI and physical examinations. We compared group-wise diffusion differences in fractional anisotropy (FA), axial diffusivity (Da) and radial diffusivity (Dr) with the voxel-based analysis (VBA).Study2:To study whether this thalamic abnormality leads to an alteration at the whole-brain level, our study investigated the role of the thalamus in patients with parkinsonian resting tremor in a large-scale brain network context. Patients with PD and healthy controls (HCs) were included in this study to undergo resting-state functional MRI scans. All the recruited PD patients were grouped into two subgroups according to the absence (NTP) and presence (TP) of resting tremor. Graph theory-based network analysis and seed-based functional connectivity analysis were performed to examine alterations in nodal centrality and functional connectivity in the three groups.Study3:Patients with PD and HCs were recruited and underwent multi-modal resting-state functional MRI scans. By comparing the PD patients with the HCs, brain features that were not conducive to the subtype-specific classification were ruled out from masses of brain features. Then, we applied a support vector machine (SVM) classifier with the recursive feature elimination (RFE) method to multi-modal MRI data for selecting features with the best discriminable power, and evaluated the proposed classifier with a leave-one-out cross-validation. We then performed a Kappa test to compare our classification results to the extant sorting scheme that used in clinical practiceResultsStudy 1:Significantly reduced FA in bilateral superior longitudinal fasciculus (SLF), bilateral anterior corona radiata and the left genu of the corpus callosum were found in the PIGD subtype compared with the non-PIGD subtype. Increased Dr in the left superior longitudinal fasciculus was exhibited in the PIGD subtype, with no statistical differences in Da were found.Study2:Compared with the HC group, patients with the TP group exhibited increased centrality in the bilateral thalami (p< 0.01), with modular structure more extensively linking to regions covering the cerebellum and motor-related cortices (p< 0.05). Moreover, significant positive correlations were found between altered centrality and tremor severity. In contrast, none of the centrality measures in the thalamus were altered in the NTP group (p> 0.39). Furthermore, our data revealed that significantly enhanced functional connectivity between the putamen and the thalamus were present in the TP group (p= 0.027 corrected for family-wise error).Study3:Using this classifier, we obtained satisfactory diagnostic rates (accuracy= 92.31%, specificity= 96.97%, sensitivity= 84.21% and AUCmax= 0.9585). The diagnostic agreement evaluated by the Kappa test showed an almost perfect agreement with the exiting clinical categorization (Kappa value= 0.83).ConclusionsStudy1:Our study confirmed previous findings that white matter abnormalities were greater in the PIGD subtype than in the non-PIGD subtype. Additionally, our findings suggested:i) compared with the non-PIGD subtype, loss of white matter integrity was greater in the PIGD subtype; ii) bilateral SLF may play a critical role in microstructural white matter abnormalities in the PIGD subtype; and iii) reduced white matter integrity in the PIGD subtype could be mainly attributed to demyelination rather than axonal loss.Study2:These findings suggest increased thalamic centrality as a promising tremor-specific imaging measure for PD, and provide evidence for the altered putamen-thalamic interaction in patients with resting tremor.Study3:With these favorable results, our findings suggested the machine learning-based classification as an alternative technique to classifying clinical subtypes in PD.