A Study On Brain Network And Prognosis Of Acute Unilateral Sudden Sensorineural Hearing Loss Based On Multimodal MRI And Machine Learning

Background Sudden sensorineural hearing loss(SSNHL)is a sudden-onset and rapidly developed hearing impairment within 72 hours,which is mostly unilateral.SSNHL is a common emergency in ENT clinic,with tinnitus,vertigo,and so on.Due to the ambiguous etiology and pathogenesis,there is few definitive studies to guide treatment.Only a few of patients can be identified with a defined cause by the routine clinical examination.Recently,some studies have showed that unilateral SSNHL is associated with the alterations in central nervous system.Human’s brain is a complex network,organized by multiple neurons or brain regions,which are functionally interactive.With the application of machine learning in neuroimaging,it’s possible to establish diagnostic/prognostic biomarkers at the individual level.Part 1.Reduced Local Segregation of Single-Subject Gray Matter Networks in Unilateral Sudden Sensorineural Hearing LossObjective: To explore topological abnormalities of single-subject gray matter networks in patients with unilateral SSNHL in the acute phase.Materials and Methods:(1)We recruited 158 patients with unilateral SSNHL and 136 age-,sex-,and educationmatched healthy controls(HC).All of them underwent pure tone audiometry(PTA)and 3D-T1 examination.(2)Individual morphological brain networks were constructed with using interregional similarity in the distribution of regional gray matter volume(GMV)after preprocessing by voxel-based morphometry(VBM).Both global(standardized clustering coefficient(g),standardized characteristic path length(λ),small-worldness(σ),clustering coefficient(Cp),characteristic path length(Lp),and global efficiency(g E),local efficiency(loc E))and nodal network metrics(nodal centrality and hubs)were calculated by graph theory analysis.(3)Independent-samples t test was performed to identify group differences in topological properties.Results:(1)The gray matter network in both unilateral SSNHL and HC were typically organized as small-world networks.Compared with HC,brain network of unilateral SSNHL patients were characterized by decreased standardized clustering coefficient(g),clustering coefficient(Cp)and local efficiency(loc E)(P<0.05).(2)Locally,unilateral SSNHL group exhibited altered nodal centrality in left superior frontal gyrus,parahippocampal gyrus,caudate nucleus,superior occipital gyrus and right lenticular nucleus(P<0.05).Altered hubs located in right superior temporal gyrus,inferior frontal gyrus and left olfactory cortex.Conclusions: The reduced segregation of network indicates individual gray matter networks in unilateral SSNHL shifted toward randomization.And the altered nodal centrality involving auditory network and other brain networks of patients with unilateral SSNHL in acute phase.Part 2.Disrupted Topological Organization in White Matter Networks in Unilateral Sudden Sensorineural Hearing LossObjective: To investigate topological organization of white matter networks in patients with unilateral SSNHL in the acute phase based on diffusion tensor imaging(DTI).Materials and Methods:(1)The study included 145 patients with unilateral SSNHL and 91 age-,sex-,and education-matched HCs.All of them underwent pure tone audiometry(PTA)and DTI examination.(2)White matter networks were constructed with using deterministic tractography.Both global(γ,λ,σ,Cp,Lp,g E,loc E)and nodal(nodal centrality and hubs)network parameters were calculated by graph theory analysis.(3)Independent-samples t test was used to evaluate group differences in topological properties.Results:(1)Both of two groups exhibited typical small-world network architecture.Compared with HC,unilateral SSNHL displayed disrupted topological organization of the white matter structural connectome indicated by significantly increased g,l,σ,loc E,Cp,g E,and decreased Lp(P<0.05).(2)Locally,unilateral SSNHL group exhibited significantly altered nodal centrality invovling auditory network,default mode network(DMN),attention network,visual network(VN)and subcortical network(P<0.05,Bonferroni corrected).Altered hubs located in auditory network,DMN,attention network,subcortical network and sensorimotor network.Conclusions: Our study revealed connectome-level alterations involved in auditory and other brain networks of patients with unilateral SSNHL in acute phase.Disrupted integration and segregation and global hypoconnectivity of white matter network might be one of potential pathophysiology of unilateral SSNHL.Part 3.Altered Functional Networks in Unilateral Sudden Sensorineural Hearing LossObjective: To evaluate abnormalities of functional network topological properties and functional connectivity(FC)in patients with unilateral SSNHL in the acute phase based on resting-state functional magnetic resonance imaging(rs-fMRI).Materials and Methods:(1)We acquired resting-state functional magnetic resonance imaging(rs-fMRI)data from 158 patients with unilateral SSNHL and 97 age-,sex-,and education-matched HCs.(2)Functional networks were constructed by calculating interregional temporal correlations.Graph theoretical analysis and independent component analysis(ICA)were applied to investigated topological properties and functional connectivity respectively.(3)Independent-samples t test and Bonferroni were performed to determine the differences of global and nodal metrics and functional connectivity between unilateral SSNHL group and HC group with age,gender and education as covariates.Results:(1)Functional networks of patients with unilateral SSNHL and HCs were typically organized as small-world networks.Compared with HC group,the global properties in functional networks showed no statistical differences.(2)Compared with HC group,the nodal properties demonstrated significantly alterations(P<0.05,Bonferroni corrected),which located in left temporal pole: superior temporal gyrus,postcentral gyrus and amygdala.Altered hubs located in left precental gyrus,postcentral gyrus and calcarine fissure and surrounding cortex.The functional connectivity between a DMN and left frontoparietal network(l FPN)decreased in unilateral SSNHL(P<0.05,Bonferroni corrected).Conclusions: The alterations of nodal metrics and FC involving auditory network,DMN and FPN in patients with unilateral SSNHL in acute phase,which could contribute to learn more about SSNHL.Part 4.Predicting Hearing Prognosis of Unilateral Sudden Sensorineural Hearing Loss via Machine Learning and Multimodal MRIObjective: With machine learning,we aim to predict the hearing outcome in unilateral SSNHL based on multimodal MRI.Materials and Methods:(1)Follow-up of the hearing outcome of patients with unilateral SSNHL 3 months later.According to prognosis,patients were divided into recovery group(n=89)and non-recovery group(n=64).(2)Support vector machine(SVM)model,Logistic regression(LR)model,and Decision Tree model were separately used for outcome prediction after features selection based on GMV,diffusional metrics(fractional anisotropy(FA),mean diffusivity(MD),axial diffusivity(AD),radial diffusivity(RD),local diffusion homogeneity(LDH)),and functional parameters(amplitude of low frequency fluctuation(ALFF),fractional ALFF(f ALFF),regional homogeneity(Re Ho)).(3)Besides,the predictive capabilities of models were evaluated by the area under ROC curve(AUC).Results:(1)AUC of training set and testing set were 0.807 and 0.752 respectively in SVM model with GMV of right supramarginal gyrus,left cingulate gyrus,right cingulate gyrus,left striatum,right striatum and left cuneus were selected.(2)Based on diffusional metrics,AUC of training set and testing set were 0762 and 0.775 respectively in LR model.After feature selection,five diffusional features remained which were LDH of right superior fronto-occipital fasciculus,LDH of body of corpus callosum, AD of right superior corona radiata,MD of left superior fronto-occipital fasciculus,and MD of left tapetum.(3)AUC of training set and testing set were 0.764 and 0.737 respectively in Decision Tree model with 6 functional parameters were selected which were ALFF of left olfactory cortex,right rolandic operculum,left caudate nucleus,left rolandic operculum,right temporal pole superior temporal gyrus and left inferior frontal gyrus,opercular part.Conclusions: Combination of machine learning and multimodal MRI,our study provides validated evidence of 17 biomarkers for predicting the hearing outcome of unilateral SSNHL,which will be helpful in clinic application.