Resting-state FMRI-based SVM Classifier:Diagnosing Insomnia Disorder And Decoding Sleep Stage

Objective:A resting-state fMRI-based support vector machine(SVM)classifier was established to automatically diagnose insomnia disorder and decode sleep stages.Methods:To recruit insomnia sufferers and healthy controls,all participants were monitored by a consecutive two-night polysomnography(PSG),And were evaluated using Pittsburgh sleep quality index(PSQI),insomnia severity index(ISI),and polysomnography(PSG),respectively.and obtained their simultaneous polysomnographic electroencephalography and functional magnetic resonance imaging(EEG-fMRI)recordings,Functional magnetic resonance imaging(fMRI)data were preprocessed by MATLAB(version 8.5R2013B)and DPABI software package,Features of functional connectivity(FC)between 116 regions were extracted based on Automatic Anatomical Labeling(AAL),Support vector machine(SVM)classification implemented in LIBSVM library with modified open-source codes classifiers were trained to characterize insomnia and decode sleep stages.Classifier performance was quantified by a 5-fold cross validation and on independent test dataset.Results:We enrolled 33 patients with insomnia disorder and 31 healthy controls,and obtained their simultaneous polysomnographic electroencephalography and functional magnetic resonance imaging(EEG-fMRI)recordings with 440 non-overlapping 5-min fMRI sessions,including 144 wakefulness-stage sessions,43 sleep Nl-stage sessions,142 sleep N2-stage sessions,111 sleep N3-stage sessions,including 218 insomnia sessions and 222 healthy sessions.(1)There was significant difference between insomnia disorder and healthy controls group on PSQI score,AIS score、latency to onset of persistent sleep(LPS)、total sleep time(TST)、Wakefulness after the onset of sustained sleep(WASO),N1 stage as a percentage of total sleep time(N1%)(P<0.05).(2)The fMRI-based SVM classifier was able to diagnose insomnia with an accuracy of 89.3%.(3)The fMRI-based SVM classifier was able to decode sleep stages with an overall accuracy of 87.3%(98.0%for stage W,56.0%for N1,85.0%for N2,and 84.0%for N3).Conclusion:We established an encouraging resting-state fMRI-based SVM classifier to automatically diagnose insomnia disorder and decode sleep stages.As an objective measure for assessing insomnia disorder,which may be used in the future.