Multimodal Studies Of Magnetic Resonance Brain Imaging In Patients With Hyperprolactinemia And Associated Pituitary Microadenomas

Part 1 Resting-state cerebral blood flow and functional network study in patients with hyperprolactinemiaObjective:To explore the changes of cerebral blood flow(CBF)at the voxel level,and functional network topological characteristics of the whole brain at resting state in HPRL patients,and possible changes of neurovascular coupling.Methods:Thirty-two patients with pathological and idiopathic hyperprolactinemia were prospectively recruited for this study.All subjects(patients and age-,sex-,and educationmatched healthy controls)underwent resting-state arterial spin labeling(ASL)and restingstate functional magnetic resonance imaging(rs-fMRI)on a 3.0 T MR scanner(Discovery MR 750,Milwaukee,WI,GE,USA).The CBF values of the HPRL group and the HC group were compared at the whole brain voxel level.Estimating topological properties of whole-brain functional networks based on the graph theory.Neurovascular coupling was represented via calculating the across-voxel correlations of cerebral blood flow(CBF)and functional connectivity strength(FCS)at brain gray matter level,and voxel-based analyses were performed on CBF,FCS,and CBF/FCS ratio maps between HPRL and HC.elemental analysis.Correlations of neuroimaging indicators with clinical variables such as PRL levels were analyzed in the patient group.Results:Compared with HC,HPRL patients had significantly lower zCBF in the right inferior frontal gyrus(involving the operculum),left middle temporal gyrus,and bilateral rectus gyrus,and significantly higher zCBF in the right middle frontal gyrus.Cp,Eg,Eloc,and synchronization were positively correlated with sparsity in both HPRL and HC groups.while Lp,Gamma,Lambda,Sigma,assortativity,and hierarchy were negatively correlated with sparsity.Both HPRL and HC groups had small-world property of their functional networks;The decreased functional network hierarchy of HPRL group was negatively correlated with serum PRL levels;the inter-module connectivity of DMN-SMN in HPRL patients decreased significantly,but no significant change in the intra-module connections of 6 modules including DMN,FPN,CON,SMN,ON and CN.The spatial distribution of CBF,FCS,and CBF/FCS ratio at the gray matter level of HPRL patients was similar to that of HC;the global CBF decreased significantly at the gray matter level,but the global FCS did not change significantly;the CBF-FCS coupling increased slightly,but the intergroup difference was not significant;the CBF/FCS ratio did not display significant intergroup changes at the voxel level of gray matter;CBF decreased significantly in the right inferior frontal gyrus and bilateral rectilinear gyrus,and increased in the right middle frontal gyrus and left superior limbic gyrus;FCS were lower in the left cingulate gyrus,left precuneus and right superior temporal gyrus,and were higer in the left inferior frontal gyrus,left middle frontal gyrus,and right middle frontal gyrus;brain regions showing significant intergroup changes of CBF did not overlap with that of FCS.There was no significant correlation between abnormal neuroimaging indicators and clinical scores such as cognition or emotion.Conclusions:Patients with HPRL showed changes CBF,FCS,and CBF/FCS coupling in several brain regions at resting-state,and the separation of functional networks was related to the abnormal increase of PRL.This study provides new insights into the neuropathophysiological mechanism of HPRL and brain microenvironment changes in the context of PRL disorders from different perspectives of cerebral blood flow,functional network,and neurovascular coupling.Part 2 Application of semiquantitative analysis of dynamic contrast-enhanced magnetic resonance imaging in pituitary microprolactinomaObjective:To investigate the diagnostic value of semiquantitative parameters of dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)in pituitary microprolactinomas.Methods:The pituitary DCE-MRI examinations of 18 patients diagnosed as pituitary microprolactinomas for the cause of HPRL in Part 1 were retrospectively collected from March 2018 to December 2022.DCE-MRI semiquantitative parameters,including relative enhancement,maximum enhancement,maximum relative enhancement,time of arrival(TO),time to peak(TTP),wash-in rate,wash-out rate,brevity of enhancement and area under the curve(AUC),of microprolactinoma lesions and normal pituitary parenchyma were measured at the same slice.Paired t-test was used to compare the parameters between microprolactinomas and normal pituitary parenchyma.Receiver operating characteristic(ROC)curves were used to analyze the diagnostic ability of semiquantitative parameters in terms of microprolactinomas.Results:Compared with the same-slice normal pituitary parenchyma,TO and TTP of microprolactinomas were significantly prolonged(all p<0.05),while relative enhancement,maximum enhancement,maximum relative enhancement,wash-in rate,wash-out rate,brevity of enhancement,and area under the curve were significantly reduced(all p<0.05).ROC analysis showed that the wash-in rate had the highest AUC value(0.941 ±0.031)and the highest specificity(0.957),which indicates its high diagnostic efficiency.The combination of wash-in rate,maximum enhancement,TTP and AUC has the highest diagnostic performance.Conclusions:Semi-quantitative analysis of DCE-MRI can reflect the perfusion changes of pituitary microprolactinomas.The semi-quantitative parameter wash-in rate has a high diagnostic efficiency for microprolactinomas.The combination of multiple parameters is beneficial to improve the diagnostic level of microprolactinomas.Part 3 Automatic segmentation of pituitary microadenomas based on dynamic contrast-enhanced magnetic resonance imagesObjective:To evaluate the artificial intelligence pituitary microadenomas segmentation models for dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI),and to explore the automatic segmentation method adapted for pituitary microadenomas..Methods:This study had two tasks.Task 1:The ResUnet and SwinUnetr models were trained by using the DCE-MRI dataset(n=174)with different data processing methods to segment the pituitary gland,and the model with better segmentation results were selected to be used for task 2.Task 2:The SwinUnetr model was trained by using the DCE-MRI dataset(n=174)with different data processing methods for the segmentation of pituitary microadenomas.The five-fold cross-validation method and Dice similarity coefficient were used for evaluating the segmentation performance.Results:For DCE-MRI images,the pituitary segmentation performance of SwinUnetr model(Dice=0.80)was better than that of ResUnet(Dice=0.71).,The segmentation performance of pituitary microadenomas was highest(Dice=0.37)when DCE-MRI images were fed into the SwinUnetr model in a 4D way.Conclusions:The SwinUnetr model based on DCE-MRI images shows good segmentation performance in the pituitary gland segmentation task,but the segmentation performance in the pituitary microadenomas segmentation task is limited and needs to be further improved.