The Early Radiation-induced Brain Injury In Patients With Nasopharyngeal Carcinoma:a Group-level To Individual-level Multimodal MRI Study

Part Ⅰ Radiation-induced surfaced-based structural and functional alterations in patients with nasopharyngeal carcinomaBackground:Radiation-induced brain injury(RBI)is an important complication in patients with nasopharyngeal carcinoma(NPC)after radiotherapy(RT).Previous MRI studies were mostly performed at voxel level.The performance of image registration was poor and the impact of the structural properties of sulcus gyrus folding on brain function was largely overlooked.Therefore,surface-based structural and functional integration analysis may help to elucide the underlying neural mechanisms of early RBIObjectives:To detect the radation-induced alterations in surface-based structural and functional metrics and structure-function integration,for uncovering the underlying neural mechanism of early RBI;Method:Forty-five healthy controls(HCs),62 Pre-RT and 45 Post-RT 0-6 months NPC patients were prospectively enrolled.Multimodel MRI data and neuropsychological data were collected.Cortical thickness and cortical surface area were calculated based on 3D T1 structural image data;surface-based regional homogeneity(2D ReHo)was calculated based on BOLD fMRI functional data.Intergroup comparison of cortical morphological indicators and functional indicators was performed,and the between-group differences in cortical morphological indicators were used as regions of interest(ROI),which was then regarded as seed for the further surface-based functional connectivity(2D FC)analysis.The correlations between cortical morphological and functional indexes,cognitive function scores and RT dosage were analyzed.Results:Surface-based cortical morphologic and functional analysis:(1)Cortical morphological analysis showed that compared with HCs,patients with NPC in pre-RT showed reduced cortical thickness in left inferior temporal gyrus(ITG);compared with HCs and pre-RT,patients with NPC in post-RT 0-6months had reduced cortical thickness in left IGL(p<0.05,permutation test-threshold free cluster enhancement,PT-TFCE correction);(2)Regional brain activity analysis showed that compared with HCs,patients with NPC in pre-RT and post-RT 0-6 months showed increased 2D-ReHo in the left ITG(p<0.05,PT-TFCE correction);compared with pre-RT,patients with NPC in post-RT 0-6 months showed increased tendency of 2D-ReHo value in the left ITG(p=0.08);(3)2D FC analysis showed that compared with HCs,patients with NPC in post-RT 0-6months showed increased 2D FC in the left ITG-right posterior central gyrus,left ITG-right precuneus(p<0.05,PT-TFCE correction);(4)Correlation analysis showed that cortical thickness in left ITG was positively correlated with MoCA score(p=0.016);2D ReHo value in left ITG was negatively correlated with MMSE score(p=0.038);2D FC value in the right precuneus was positively correlated with MDRT of the right temporal lobe(p=0.030).Conclusion:Radiation-induced structural and functional alterations could be observed in the left ITG at post-RT 0-6 months in patients with NPC.The abnormal functional integration between the left ITG and somatosensory system as well as default network-related brain regions suggested that RBI may be a multisystem disorder that involves intratemporal and extratemporal lobe regions.Part Ⅱ An fMRI study of radiation-induced neurovascular decoupling in prediction of cognitive decline in patients with nasopharyngeal carcinomaBackground:Radiation-induced vascular damage may be an important factor of cognitive impairment in patients with NPC;however,how does the cerebrovascular damage affect brain function changes and thus affect the cognitive function in NPC is still unclear.Therefore,exploring the radiation-induced cognitive impairment from the perspective of neuro-vascular coupling may be a new breakthrough for this issue.Objectives:To study the neural basis of radiation-induced neuro-vascular de coupling in patients with NPC,and to explore the effect of neuro-vascular decoupling on cognitive function impairements after RT;Method:Sixty-two pre-RT and 45 post-RT 0-6 months patients with NPC with multimodal MRI data and neuropsychological data were prospectively collected.The fractional amplitude of low-frequency fluctuation(fALFF)was calculated based on BOLD fMRI data;cerebral blood flow(CBF)was calculated based on 2D ASL data.After that,the fALFF/CBF was obtained to assess neuro-vascular coupling.Using fALFF/CBF,a support vector machine(SVM)algorithm based predictive model was constructed to predict cognitive impairment after RT.Results:Neuro-vascular decoupling analysis:(1)Between-group comparison showed that compared with patients with NPC in Pre-RT,patients in post-RT 0-6months showed decreased fALFF/CBF in bilateral posterior cerebellar lobe,left anterior cerebellar lobe and tonsil,and increased fALFF/CBF in the bilateral putamen and globus pallidus,left insula,left anterior cingulate gyrus and left medial prefrontal lobe(Alphasim correction);(2)Cognitive dysfunction prediction modal:The fALFF/CBF of bilateral temporal lobes(including inferior temporal gyrus,fusiform gyrus,etc.),medial and lateral prefrontal lobes,right superior marginal gyrus,right inferior parietal lobule,and right cuneus/occipital lobe contribute to the prediction modal construction of radiation-induced cognitive impairmen.The area under the ROC curve(AUC)of the proposed model was 0.89,and the accuracy was 0.81.Conclusion:The fALFF/CBF values inside and outside the radiation fields were altered in the opposite direction,and it could effectively predict radiation-induced cognitive impairment,suggesting that neuro-vascular decoupling may be the neural basis underlying cognitive decline in RBI.Part Ⅲ A multi-center study of multimodal MRI basedRadiomics in prediction of radiation-induced brain necrosis in patients with nasopharyngeal carcinoma Background:Limited radiomics studies have demonstrated the feasibility of using multimodal brain imaging features after RT to predict the occurrence of radiation necrosis(also named radiation encephalopathy,RE);however,the small sample size from single-center limits the robustness and generalizability of the prediction models;therefore,a prediction study using larger sample,serial multimodal MRI data from multi sites is essential for the early diagnosis of RE.Objectives:To construct a RE prediction model based on multi-center and large-sample MRI data,and to perform risk stratification analysis for RBI patients.Method:Eight hundred and eighty five patietns with NPC with multiple time points(including Pre-RT and Post-RT 0-3 months)multimodal MRI data(including T1 WI,T1+C,T2WI)from multi-center were retrospectively collected.Bilateral temporal lobes and/or gray and white matter,and brainstem were manually annotated on each MRI modality.Feature extraction and selection,as well as model training and validation are performed using the radiomics research framework.External dataset is used to test the proposed model.The performance of each predictive model was calculated.Stratified analysis of the risk of radiation necrosis after RT for NPC was performed accoreding to the output of the proposed model.The correlations between Radscore and post-radiotherapy serological data as well as RT dosage were calculated.Results:Multi-center individualized prediction:(1)single modality based RE prediction model showed that the performance of T1+C-model was higher than that of T2 WI and T1 W].The AUCs of pre-RT-T1+C-model,pre-RT-T2 WI-model and pre-RT-T1 WI-model were 0.914,0.760,0.830(test set);The AUCs of post-RT 0-3months-Tl+C-model,post-RT 0-3months-T2 WI-model and post-RT 0-3months-T1 WI-model were 0.975,0.950,0.622(test set),respectively;(2)The performance of pre-and post-RT 0-3 months combined model was better than that of pre-RT-model or post-RT 0-3 months-model.The AUCs of the pre-RT and post-RT 0-3 months combined model,the pre-RT-model,and the post-RT 0-3 months model were 0.920,0.908,0.914(test set),respectively;(3)the integration of imaging features of temporal lobe gray and white matter can improve predictive model performance;(4)Radscore could effectively classify patients with NPC into high-,intermediate-and low-risk subgroups(p<0.0001);(5)Radscore was positively correlated with mean radiation dose of the right temporal lobe(p=0.024),and negatively correlated with serum concentration of HDL-C(p=0.029)and APOA1(p=0.030).Conclusion:The pre-and post-RT 0-3 months combined radiomics model could effectively predict the occurrence of RE,which could further help to make risk stratification of NPC patients.These findings may contribute to achive individualized treatment in paients with NPC.