Multi-modal Magnetic Resonance Imaging And Artificial Intelligence In Patients With Stroke

Part ? The length and offset varied saturation magnetic resonance imaging in the patients with acute ischemic stroke and acute intracerebral hemorrhageObjective:To explore the value of the length and offset varied saturation(LOVARS)magnetic resonance imaging(MRI)in detection and separation of acute ischemic stroke and acute intracerebral hemorrhage.Methods:11 patients with acute ischemic stroke and 9 patients with acute intracerebral hemorrhage within 24 hours after symptom onset were enrolled in this study.Routine MRI(T1WI,T2WI,FLAIR and DWI)and LOVARS MRI were performed among all patients on a 3.0 T MR scanner(Magnetom Trio,Siemens Medical Solutions,Erlangen,Germany)with a 12-channel head coil.All the patients with ischemic stroke were confirmed by CT or DWI and all the patients with intracerebral hemorrhage were confirmed by CT or SWI.Two radiologists independently drew the region of interest(ROI)of the lesion on each map for the statistical analyses.The ROIs contained the entire abnormal areas on LOVARS MRI real images and the contralateral normal tissues.The independent-samples t test and paired-samples t test were performed to compare the LOVARS MRI real signal intensities among the acute intracerebral hemorrhage lesions,acute ischemic stroke lesions and the respective contralateral normal white matters,and P<0.05 was considered statistically significant.Results:The patients with acute intracerebral hemorrhage had significantly higher LOVARS MRI real signal intensities than the patients with acute ischemic stroke.The LOVARS MRI real signal intensities were significantly higher in the acute intracerebral hemorrhage lesions and lower in the acute ischemic stroke lesions than that in the respective contralateral normal white matters(all P<0.05).Conclusion:LOVARS MRI could simultaneously detect and separate the acute intracerebral hemorrhage and acute ischemic stroke.Part ? Absolute quantitative dynamic susceptibility contrast cerebral perfusion imaging using the self-calibrated echo planar imaging sequence in patients with ischemic strokeObjective:An absolute quantitative dynamic susceptibility contrast(DSC)cerebral perfusion weighted imaging(PWI)technique based on the self-calibrated echo planar imaging sequence(SCALE-PWI)could be the reliable measurement of quantitative cerebral blood flow(qCBF)and quantitative cerebral blood volume(qCBV)in a short scan time.The purpose of this study was to investigate the clinical value of SCALE-PWI in detecting salvageable areas in the patients with ischemic stroke.Methods:Fourteen patients with ischemic stroke underwent routine MRI(T1WI,T2WI,FLAIR,DWI)and SCALE-PWI scanning on a 3.0 T MR scanner(Prisma MR,Siemens Healthcare,Erlangen,Germany)with a 64-channel head coil.The subjects were scanned with a delay of 46 s between consecutive modules of SCALE-PWI.A single dose(0.1 mmol/kg body weight)of Gd-DTPA contrast agent(Magnevist,Berlex,Montville,NJ,USA)was injected during the DSC module of SCALE-PWI,followed by a 20 mL saline flush,at a rate of 4 mL/s.The qCBV and qCBF maps were inline processed immediately after the scan.Two radiologists independently drew the ROIs of the lesion on each map for the statistical analyses.The ROIs contained the entire abnormal areas on the qCBV and qCBF images,the lesions that appeared hyperintense on DWI,the DWI/PWI mismatch areas and the contralateral normal tissues.Paired t-tests,Wilcoxon signed-rank tests and receiver operating characteristic curve were performed,and P<0.05 was considered statistically significant.Results:All the values of qCBF and qCBV in the lesions were lower than that of the contralateral normal tissues(all P<0.05).The values of qCBF and qCBV in the ischemic core were lower than that of the penumbra(mean value:16.42 mL/100g/min vs.21.54 mL/100g/min,P=0.013;1.23 mL/100g vs.1.47 mL/100g,P=0.049;respectively).The threshold qCBF of ischemic core was 18.18 mL/100g/min and the threshold qCBF of penumbra was 28.09 mL/100g/min.Conclusion:Different from the previous semi-quantitative method,the SCALE-PWI technique could provide absolute quantitative hemodynamic information which could be used to detect the ischemic core and penumbra in a relatively short scan time,and thus may serve as a guide for tissue-based decision-making and personalized treatment planning in acute ischemic stroke.Part ? Deep learning tool for acute ischemic stroke in volumetric segmentation and outcome predictionObjective:To develop a deep learning tool for ischemic stroke to segment ischemic core volume and predict outcome of patients with acute ischemic stroke.Methods:The ischemic core lesions on MR DWI from 1205 consecutive patients with acute ischemic stroke were manual annotated.Use the manual annotations to develop and train a deep convolutional neural network.The modified U-Net neural network architecture and ensembled output of several trained models were used to improve the segmentation.Testing set S(?)rensen-dice similarity coefficient with the Mann-Whitney U test was used to evaluate the performance of individual models and ensembled models,and the volumes between the different mRS stratifications.Intraclass correlation coefficient(ICC)was used to evaluate the relatedness between manual segmentation volume and automatic segmentation volume.P<0.05 was the threshold for all statistical tests.Results:The ensemble of 4 Inception 3D U-Nets performed best(median S(?)rensen-dice similarity coefficient=0.737).There was a high agreement between manually and automatically derived volumes(ICC=0.977).For independent predictors,the algorithm performed less well in punctate infarcts(median volume=0.155 mL)and in infarcts exclusively located within the cortex or deep white matter.The automatic volume had potential to predict the outcome.Conclusion:This fully automatic segmentation tool based on deep learning could accurately segment infarction core and has broad clinical application and scientific prospects.