| Background and purpose:Gliomas was the most common primary epithelial tumors in adult brain,accounting for more than 70% of primary tumors in the central nervous system.In recent years,the incidence of glioma has shown a significant upward trend.According to CBTRUS statistics,about 5.26/100,000 adults were diagnosed with glioma every year between 2008 and 2012 and according to data released by EANO and EORTC,the annual incidence of glioma in the European population is 6/100,000 by 2017.The fourth edition of WHO Classification of Central Nervous System Tumors in 2007 and its update in 2016 classified gliomas into four pathological grades: I,II,III and IV based on their histomorphological characteristics.The pathological grade of glioma was closely related to the biological behavior of glioma.The higher the grade of glioma,the more malignant it was.The corresponding clinical intervention strategies were also different.After long-term follow-up,the RTOG 98-02 trial recommended that high-grade gliomas should be treated as early as possible after surgery.However,the optimal timing and risk of radiation neurotoxicity for low-grade gliomas remain controversial.The prognosis of glioblastoma was poor and the recurrence rate was high.Despite active surgical treatment combined with radiotherapy and chemotherapy,the 5-year survival rate of glioblastoma patients was still less than 10%.There were two main ways of recurrence,local and distant.A large number of autopsy results showed that although GBM grew aggressively in the brain,most of the tumors recurred locally.Many results showed that the local recurrence rate of GBM accounts for 80% or even higher of the total recurrence rate.Pope et al.followed up 110 GBM patients and found that distant recurrence was closely related to poor prognosis.Predicting GBM recurrence pattern before operation was of great clinical significance in the selection of individualized radiotherapy schemes for various recurrence patterns.Tumor neovascularization was one of the important milestones in the evolution of glioma and anti-angiogenic therapy was an important supplement to conventional chemotherapy for glioma after operation and recurrence.Neovascularization was a complex pathophysiological process in glioma,including vascular co-option,angiogenesis,vasculogenesis,mosaic vessel formation,vascular mimicry,glioblastoma-endothelial cell transdifferentiation and intussusceptive microvascular growth(IMG),tumor cells and vascular endothelial cells participated in and interact with each other to synergistically influence angiogenesis of glioma.Anti-angiogenic therapy targeting the key molecules of angiogenesis can inhibit angiogenesis of a glioma.However,clinical evidence showed that most of the anti-angiogenic strategies were far from expected,and the long-term outcomes of some patients may even lead to malignant progression of tumors.JAK/STAT signaling pathway played a leading role in cytokine signaling.Signal transducers and activators of transcription(STAT3),as a key molecule in JAK/STAT signal transduction pathway,was a key regulator in various core events of cancer.Large amounts of experimental evidences showed that STAT3 was continuously activated in tumor cells and endothelial cells.Evidence has shown that inhibiting STAT3 activity can effectively reduce vascular mimicry and vascular symbiosis in glioma.However,the relationship between STAT3 activity and intussusception angiogenesis has not been reported.If there was evidence to clarify the relationship between STAT3 and IMG,it can be confirmed that STAT3 was widely involved in various forms of glioma neovascularization,thus providing a strong experimental basis for STAT3 to become a new and "broad-spectrum" target for anti-angiogeneic therapy.Radiomics collected image information,including CT,MRI and PET,and analyzed the corresponding image features of the lesion area,and then described the overall situation of the tumors in detail with the big data imaging methods.The quantitative semantic VASARI feature set based on image visual features designed by TCIA defined a set of reliable descriptive image features.Studies have shown that VASARI features can not only be used to describe the morphology of glioma,but also have potential value in predicting the expression of characteristic molecules,molecular subtypes and prognosis of glioma.Lee et al.analyzed preoperative MRI images of 86 patients with anaplastic astrocytoma and extracted 8 VASARI features related to Overall Survival(OS)and Progression-free Survival(PFS)to establish a radiological risk score(RRS)for predicting the prognosis of patients.VASARI needed further evaluation in judging the type of GBM recurrence and evaluating drug response.Based on this,the aim of this study is to evaluate the clinical strategies of glioma before operation by using conventional MRI techniques.(1)By extracting and quantifying the "semantic features" and "unknown features" of glioma,stable and reproducible image features were analyzed and screened out.(2)Preoperative pathological grading of glioma patients was analyzed by imaging combined with clinical baseline data and post-operative pathological grading.(3)Different patterns of GBM recurrence were summarized according to the imaging characteristics.Progressive free survival(PFS)was followed up and the prognosis of GBM was evaluated by quantitative imaging.(4)To explore the correlation between STAT3 and GBM angiogenesis,to clarify the potential value of STAT3 in anti-angiogenesis strategy of GBM,and then to predict the expression of STAT3 by quantitative imaging features before operation,so as to predict the anti-angiogenesis response of GBM in advance.Through the above four aspects of clinical strategy evaluation,we hope to provide valuable theoretical support for the selection of imaging strategies for evaluating clinical events of glioma patients based on conventional MRI.Materials and methods:Following the above research objectives,this study consisted of four parts,including " Quantitative MRI feature extraction and stability analysis of glioma"," Preoperative grading of glioma based on quantitative VASARI features","Retrospective study on recurrence patterns and prognosis of GBM based on quantitative VASARI features" and "MRI analysis of STAT3 expression and angiogenesis and prognosis of glioblastoma ".The first part1.Research object: The preoperative MRI data of 153 patients with glioma confirmed by histopathology in the Third Affiliated Hospital of Army Medical University from June 2014 to September 2016 were retrospectively collected.2.Acquisition of preoperative MRI data of glioma patients: Preoperative MRI data of 153 patients with glioma were collected by 3.0T magnetic resonance imaging.The scan sequence included T1 W,T2W,FLAIR,DWI and T1 W enhancement.3.Extraction of quantitative imaging features:3.1 Quantitative VASARI semantic features were independently extracted by three radiologists with more than five years’ experiences.3.2 Three radiologists with 5 years or more working experience independently segmented the four tumor regions of glioma: enhancement,necrosis,edema and tumor body.Quantitative imaging features such as "Intensity","Texture" and "Shape" were extracted by IBEX toolkit.4.Statistical analysis: Cohen kappa analysis of the stability of quantitative imaging features of VASARI.Those with a kappa value less than 0.5 were considered to be basically unstable,those with a stability of 0.5 < kappa < 0.8 were identified as "low",those with a stability of 0.8 < kappa < 0.9 were identified as "medium",and those with a stability of kappa > 0.9 were identified as "high".ICC was used to check the stability of quantitative imaging features based on gray distribution.ICC > 0.9 was defined as high stability,ICC <= 0.5 as low stability,and 0.5 < ICC < 0.9 as medium stability.The second part1.Research Object: The preoperative MRI and clinical data of 153 patients with glioma confirmed by histopathology in the Third Affiliated Hospital of Army Medical University from June 2014 to September 2016 were retrospectively collected.2.Acquisition of preoperative MRI data of glioma patients: Preoperative MRI data of 153 patients with glioma were collected by 3.0T magnetic resonance imaging.The scan sequence included T1 W,T2W,FLAIR,DWI and T1 W enhancement.3.Extraction of quantitative imaging features: Quantitative VASARI semantic features were independently extracted by three radiologists with more than five years’ experiences.4.Statistical analysis: Cohen Kappa analysis was used to analyze the stability of quantitative characteristics.It was found that the stability of quantitative characteristics was higher when the value of kappa was greater than or equal to 0.5,and the stability was higher when the value of kappa was less than 0.5.Kruskal-Wallis H test was used to screen quantitative features with statistical differences among gliomas of different grades.Binary logistic regression was used to validate the quantitative characteristics with statistical differences and to test the effectiveness of grading diagnosis of glioma.The third part1.Research Object: Preoperative MRI and clinical data of 65 patients with recurrent GBM diagnosed in the Third Affiliated Hospital of Army Medical University from June 2014 to September 2016 were retrospectively collected.2.Preoperative MR data of recurrent GBM patients were collected: 3.0T MR imager was used to collect preoperative MR data,including T1 W,T2W,FLAIR,DWI and T1 W enhancement.3.Extraction of quantitative imaging features: Quantitative VASARI semantic features were independently extracted by three radiologists with more than five years’ experiences.4.Follow-up and analysis of recurrence patterns: PFS was recorded.According to the location of new lesions on post-operative T1 W enhanced images,recurrence types were divided into "local recurrence" and "distant recurrence".5.Statistical analysis: Mann-Whitney U test was used to screen VASARI features with significant differences between the two recurrence patterns.The difference of PFS in recurrent GBM patients was analyzed by Kaplan-Meier curve.Cox proportional risk model was selected for multivariate analysis,and risk factors influencing PFS with statistical significance were screened.The fourth part1.Research Object: The preoperative MRI and clinical data of 27 patients diagnosed as GBM by histopathology in the Third Affiliated Hospital of Army Medical University from June 2014 to September 2016 were retrospectively collected,and OS was recorded.2.Acquisition of preoperative MRI data of glioma patients: Preoperative MRI data of 27 patients with glioma were collected by 3.0T magnetic resonance imaging.The scan sequence included T1 W,T2W,FLAIR,DWI and T1 W enhancement.3.Extraction of quantitative imaging features: Quantitative VASARI semantic features were independently extracted by three radiologists with more than five years’ experiences.4.Histopathological examination of GBM: Microvascular MVA,VSI,Sprout,IMG and STAT3 were quantified by IHC.5.Statistical analysis: Chi-square test was used to compare the correlation between STAT3 expression and clinical indicators,and Spearman was used to compare the correlation between neovascularization indicators.Log-rank test was used to examine the correlation between STAT3 and OS in patients with GBM.Kaplan-Meier curve was used to draw survival curve of GBM.Cox proportional risk model,screening the risk factors affecting OS.Z test screened VASARI features with significant difference in STAT3 expression.ResultsThe first part1.Stability of VASARI feature set1.1 A total of 30 quantitative imaging features were extracted from the VASARI feature set.1.2 All 30 features have good stability,the lowest stability was F21(Deep WM invasion),and the highest stability was F29(Lesion Size).1.3 There were 12 features with "low" stability,such as F3(Eloquent Brain),F4(Enhancement Quality),F7(Proportion Necrosis),F8(Cyst),F11(Thickness of enhancing margin),F17(Diffusion),F19(Ependymal invasion),F20(Cortical involvement),F21(Deep WM invasion),F22(nCET tumor Crosses Midline),F27(Extent resection of nCET)and F28(Extent resection of vasogenic edema);12 features with "medium" stability,such as F1(Tumor Location),F5(Proportion Enhancing),F6(Proportion nCET),F9(Multifocal or Multicentric),F10(T1/FLAIR Ratio),F12(Definition of the enhancing margin),F13(non-enhancing margin),F15(Edema Crosses Midline),F16(Hemorrhage),F18(Pial invasion),F24(Satellites)and F26(Extent of resection of enhancing tumor);6 features with "high" stability were defined as F2(Side of Tumor Epicenter),F14(Proportion of Edema),F23(Enhancing tumor Crosses Midline),F25(Calvarial remodeling),F29(Lesion Size)and F30(Lesion Size),respectively.2.Stability of quantitative image features based on gray level2.1 According to the gray level feature set,three feature sets and 401 quantitative imaging features were extracted,and most of the stability can be defined as "medium" or above.2.2 The stability among different MRI scanners was "moderate";the stability of "Shape" feature set in three feature sets was relatively poor;and the data stability of "edema" area in four ROIs was relatively low.The second part1.There were significant differences in 10 quantitative characteristics of gliomas in different grades,including F3(Eloquent Brain),F4(Enhancement Quality),F5(Proportion Enhancing),F6(Proportion nCET),F7(Proportion Necrosis),F11(Thickness of enhancing margin),F12(Definition of the enhancing margin),F17(Diffusion),F19(Ependymal invasion)and F22(nCET tumor Crosses Midline)(P<0.001).2.Eloquent Brain(F3)was of great value in differentiating between high-grade and low-grade gliomas and between WHO grade II and III gliomas.3.nCET tumor Crosses Midline(F22)was of great diagnostic value in predicting grade IV glioma.The third part1.Of 65 cases of recurrent GBM,49 cases had local recurrence and 16 cases had distant recurrence.2.There were significant differences in the three VASARI features of non-enhancing margin(F13),cortical involvement(F20)and deep white matter invasion(F21)between the two groups.3.F13(non-enhancing margin)was an important predictor of poor PFS.The fourth part1.The expression of STAT3 in GBM tissues was significantly different,while MVA and IMG were positively correlated with STAT3 expression.2.Univariate analysis showed that IMG,MVA and STAT3 were significantly correlated with the prognosis of GBM.Multivariate Cox analysis revealed that high expression of STAT3 was an independent predictor of poor prognosis of GBM.3.Three VASARI features,necrosis proportion(F7),non-enhancing margin(F13)and cortical involvement(F20),were different among patients with different STAT3 expression.Conclusion1.Quantitative image features based on conventional magnetic resonance imaging technology include VASAR semantic features and quantitative features based on gray distribution,both of which have good stability.2.Quantitative semantic features of VASARI were valuable in preoperative identification of high-and low-grade gliomas,WHO II and III,and WHO III and IV gliomas.3.Recurrence of GBM can be achieved locally and remotely.And quantitative semantic features of VASARI can differentiate them.At the same time,VASARI features were also related to the poor prognosis of recurrent GBM.4.The expression of STAT3 in GBM tissues was different,and may affect the prognosis of patients by participating in tumor angiogenesis.VASARI features were valuable in predicting the expression of STAT3 before operation.5.Quantitative VASARI features based on conventional magnetic resonance imaging technology had certain value for preoperative grading,prognosis evaluation and therapeutic effect detection of glioma,and were expected to provide more valuable information i n the decision-making process of clinical events of glioma. |
PDF Full Text Request