MR T1? Imaging And Brain Structure MR Imaging On Radiation-induced Brain Injury In Patients With Nasopharyngeal Carcinoma&MR Texture Analysis On Predicting Recurrence Of Nasopharyngeal Carcinoma

Part One MR T1? imaging and brain structure MR imaging on radiation-induced brain injury in patients with nasopharyngeal carcinomObjective:To analysis microcosmic changes of structure and morphology on normal-appearing brain in patients with nasopharyngeal carcinoma(NPC)after radiotherapy(RT)by MR T1? imaging and brain structure MR imaging,and to explore the diagnostic value of these new MR technologies on early diagnosis of latent radiation-induced brain injury in NPC patients after RT.Materials and methods:1.Research objects138 patients(including 102 men and 36 women with an average age of 46.58 years)with NPC(clinical stages ranged from T1N1M0 to T4N3M0)who had been diagnosed by histopathology were enrolled.The patients were classified into four groups on the basis of the MR examination time before and after completion of RT.There were pre-RT group(control group,38 cases),0?6 months post-RT group(40 cases),>6?12 months post-RT group(32 cases)and>12 months post-RT group(28 cases).2.MR images acquisitionAll MR scans were performed on Philips Achieva 3.0T MR scanner.Conventional MR plain scanning included axial T1-weighted imaging,T2-weighted imaging,FLAIR sequence and sagittal T1-weighted imaging.Enhancement scaning contained axial,coronal and sagittal T1-weighted imaging.T1? imaging was performed using 3D Turbo Spin-Echo pulse sequence,with spin lock frequency of 500 Hz and spin lock time of 0,20,40,60,80 and 100 ms respectively.The entire volume of each subject's brain was imaged in the sagittal plane using a T1-weighted 3D volumetric pulse sequence.3.T1? imaging post-processingThe main steps are listed as follows:(1)Format conversion.(2)Non-brain tissues were removed.(3)Correction.(4)Registration,segmentation and normalization.(5)Extraction of brain regions in cortical gray matter(GM)and white matter(WM)of each subject.(6)Extraction of T1? values in each brain region of each subject.4.Brain structure MR imaging post-processingThe main steps are listed as follows:(1)Format conversion.(2)Non-brain tissues were removed.(3)Non-linear registration,non-linear transformation and inversion.(4)The AAL template was deformed by inverse transformation to obtain the brain regions of each subject.(5)Generate the number of voxels in each brain region of each subject.5.Statistical analysisStatistical analysis was performed using SPSS version 20.0 or MATLAB2015a workstation.The measurement data were recorded as mean±standard deviation.The statistically significant differences was set at p<0.05.Chi-square Test was performed to analyze the difference of gender in NPC patients of all groups.One-way ANOVA analysis was used to compare the difference of age and T1? values of GM and WM in NPC patients of all groups.Permutation Test was performed to analyze the number of voxels in NPC patients of all groups.The pre-RT group was defined as the control group,and all post-RT groups were compared with control group respectively,with 10000 times' operation.Results:1.There were no significant differences in age and gender of all groups.2.Compared with NPC patients of pre-RT group,the T1? values of GM and WM in NPC patients of post-RT groups showing a dynamic changing process in different periods after RT,with a trend from rise to fall and finally rise,or a trend of continuous rising.3.Compared with NPC patients of pre-RT group,there was significantly increase or decrease in the volumes of multiple brain regions in NPC patients of post-RT groups in different periods after RT.Conclusions:1.MR T1? imaging and brain structure MR imaging are sensitive to monitor the microcosmic alterations of structure and morphology in normal-appearing brain of NPC patients after RT,indicating that these new MR technologies may provide useful imaging tools for early diagnosis on latent radiation-induced brain injury.2.The T1? values of GM and WM in NPC patients show a dynamic change process in different periods after RT.The volumes of multiple brain regions in NPC patients also show increase or decrease in different periods after RT.All these indicate that RT could cause extensive brain damage,not only in the irradiation field.Furthermore,the response of brain tissue to radiation is intricate and dynamic.3.The rise of T1? values in NPC patients of 0?6 months post-RT group or>12 months post-RT group might be related to neuronal and glial cell damage or fibrosis.The fall of T1? values in NPC patients of>6?12 months post-RT group might reflects the repair effect of radiation-induced brain injury.The rise of T1? values in NPC patients of>6?12 months post-RT group might reflects the progress of(?)brain injury or suggesting that the damage effect is much greater than the repair effect.4.The reduction in the cerebrum volume of multiple regions of NPC patients in different periods after RT might be related to radiation-induced neuronal and glial cell damage,vascular injury,autoimmune response or chemotherapy.The increase in the whole brain volume of multiple regions of NPC patients in different periods after RT might be a compensatory response secondary to radiation-induced brain injury.Part Two MR texture analysis on predicting recurrence of nasopharyngeal carcinomaObjective:To analysis the texture features of MR images in nasopharyngeal carcinoma(NPC)patients retrospectively,and to evaluate the value of radiomics model based on MR texture analysis in predicting recurrence of NPC.Materials and methods:1.Research objects85 patients with NPC who had been diagnosed by histopathology were enrolled.The patients were divided into two groups according to their prognosis.(1)Recurrent NPC group:35 cases,including 28 men and 7 women with an average age of 45.77 years.The clinical stages ranged from TiNiMo to T4N3M0.(2)Nonrecurrent NPC group:50 cases,including 34 men and 16 women with an average age of 46.50 years.The clinical stages ranged from T2N0M0 to T3N3M0.2.MR images acquisitionAll MR scans were performed on Siemens Vision Plus 1.5T and GE Signa Excite 3.0T MR scanner.MR plain scanning included axial T1-weighted imaging,T2-weighted imaging,coronal T2-weighted imaging and sagittal T1-weighted imaging,and enhancement scanning contained axial,coronal and sagittal T1-weighted imaging.3.Image texture analysis(1)Images selection.The nasopharynx MR images of all NPC patients before treatment were exported from the PACS system.(2)Volume of interest selection.The axial enhanced T1-weighted images and T2-weighted images of NPC patients were imported in the ITK-SNAP software.The tumors were outlined manually.(3)Images pre-processing.Including offset field correction and gray scale normalization.(4)Features extraction.?4 non-texture features.?43 texture features.?Selection of texture extraction parameters.(5)Features selection and classification.Feature selection via eigenvector centrality(EC)algorithm and Feature selection via concave minimization(FSV)algorithm were adopted to select texture features.Bootstrap method and Random forests algorithm were used to classify the selected features.The predictive effect of the two texture features selection methods was analyzed and compared by using the receiver characteristic curve.(6)Radiomics model establishment and prediction effect evaluation.4.Statistical analysisStatistical analysis of clinical data was performed using SPSS version 20.0.The measurement data were recorded as mean ± standard deviation.The statistically significant differences was set at p<0.05.Independent-samples T testwas used to compare the difference of age between recurrent NPC group and nonrecurrent NPC group.Chi-square Test was performed to analyze the difference of gender and clinical stages between the two groups.Statistical analysis of image texture analysis was performed using related machine-learning algorithms in MATLAB2015a workstation.Results:1.There were no significant differences of age,gender and clinical stages between recurrent NPC group and nonrecurrent NPC group.2.For recurrence of NPC,the prediction effect of FSV algorithm was obviously better than that of EC algorithm.3.For enhanced T1-weighted images,20 texture features combination with FSV algorithm showed the best predictive effect.The model reached an area under the receiver operating characteristic curve(AUC)of 0.866,a sensitivity of 75.10%,and a specificity of 86.56%.For T2-weighted images,15 texture features combination with FSV algorithm showed the best predictive effect.The model reached an AUC of 0.843,a sensitivity of 73.49%,and a specificity of 83.11%.4.The best performance was obtained using a combination of 20 texture features extracted from enhanced T1-weighted images and 15 texture features extracted from T2-weighted images with FSV algorithm.This model reached an AUC of 0.878,a sensitivity of 75.73%,and a specificity of 87.11%.Conclusions:1.Texture analysis based on MR images could predict the recurrence of NPC effectively,indicating that MR texture features can be used as biomarkers to predict the recurrence of NPC.2.Radiomics is expected to become a new radiological analysis tool for early prognosis evaluation of NPC patients.