MRI Texture Analysis For Differential Diagnosis Of Cerebral Glioma Following World Health Organization(WHO) 2016 Classification Of Central Nervous System(CNS) Tumors

Purpose: to determine the strength of MRI texture analysis using conventional sequences to distinguish cerebral gliomas grade II,III and IV by taking in account their genetic status in accordance with the 2016 Classification of CNS tumours and define probable correlation between textural parameters and patho-genetic changes;thus,elaborate a list of potential textures features that might be routinely used for the accurate non-invasive diagnostic and treatment planning of cerebral gliomas.Materials and method: 51 patients(27 females,24 males)with cerebral gliomas were collected in this retrospective study,which included: 15 cases of WHO II gliomas,19 cases of WHO III,and 17 cases of WHO IV.Axial T1 WI,T2WI and contrasted T1 WI MRI DICOM(Digital Imaging and Communication in Medicine)format images performed with GE 1.5 Tesla MR Scanner(Signa HDxt,General Electronic Co,Milwaukee,WI,USA)were saved using ADW4.6 Workstation.The MR texture analysis(MRTA)was performed with OMNIKINETICS Software(GE Healthcare,CHINA)and then the 3D ROIs were delimited manually on all axial slices covering the entire tumor parenchyma.;Basically five groups were defined: the first group that divided the tumors into grades(WHO II-III-IV),the other four groups were defined according to the presence or lack of mutation and the absence of test to a specific genetic mutation(P53,IDH-1,GFAP,EGFR);in other terms: we compared patients tested and positive to a mutation,patients tested and negative to a mutation and patients who weren't tested for that same mutation.The percentage of Ki-67 was then compared with the parameters showing enough statistical degree to differentiate the parts of each defined group and on each MRI sequences to check potential existing relation betweenthem.First order(Min Intensity,Max Intensity,Mean Intensity,Uniformity,Entropy,Standard Deviation,Skewness,Kurtosis and Quantiles)and second-order(based on cooccurrence matrix: GLCM-Entropy,energy,Angular Second Moment,Homogeneity,Correlation,and Dissimilarity and based on Run Lenght Matrix: Short Run Emphasis(SRE),Long Run Emphasis(LRE),Grey Level Non-Uniformity(GLNU),Run Length NonUniformity(RLNU),Short Run Low Grey Level,Long Run Low Grey Level,Short Run High Grey Level and Long Run High Grey Level)texture parameters of T1 WI,T2WI and contrasted-T1 WI were measured and selected for statistical analysis which was done on SPSS 24 software;One-way ANOVA and Kruskal-Wallis test were used to define parameters that can differentiate each part of a group.Receiver Operating Characteristics(ROC)curve was used to assess sensitivity and specificity to differentiate our tumors' types that were compared according to their grades and their genetic mutations(TP53,IDH-1,EGFR,and GFAP).Spearman-rank correlation permitted to determine the relationship between significant(statistically)texture parameters and the percentages of Ki-67.p < 0.05 symbolized statistical significance.Results: 51 patients were included: 27 females and 24 males with the mean age=21.78±21.13 years.The mean-percentage of Ki-67 estimated 50.1% with maximum of 72% and minimum of 27%.In total,27 patients had IDH-1 mutation,30 TP53 mutation,33 GFAP mutation and 10 EGFR mutation.Grades II,III and IV gliomas were differentiated on all sequences,mostly on contrasted-T1 WI where were recorded the highest AUC=0.858 corresponding to Skewness and the highest sensitivity=100% corresponding to Cluster Shade(AUC=0.825);the Highest specificity=83% was obtained on T2 WI and corresponded to Variance(P=0.009,AUC=0.739).On T1 WI,Short Run Low Grey Level(SRLGL,p=0.033,AUC=0.670)was significant with specificity of 50% and sensitivity of 80%.The group P53-mutation was differentiated on T2 WI with highest AUC=0.745,sensitivity=92% corresponding to Skewness.Cluster Shade recorded highest specificity=72%.The parameter Inertia had the highest AUC and Sensitivity on both contrasted-T1WI(0.828,92%)and T2WI(0.832,100%)while differentiating the IDH-1 mutation's group.The highest specificity=83% occurred on T1 WI pre-contrast and corresponds to Uniformity.For GFAP-mutation,the specificity and sensitivity reached 100% on all sequences.The highest AUC= 0.935,0.857 and 0.794 recorded on T1 WI,T2WI and T1WI+C,respectively and corresponded to parameters Contrast,Min-Intensity and Quantile5.EGFR-mutation group was well differentiated on T2 WI with highest AUC=0.837,Sensitivity=100% corresponded to parameter Inertia;Cluster Prominence recorded highest specificity of 100%.On T1 WI Quantile5(p=0.032)recorded highest AUC=0.748,specificity=75% and sensitivity=72%.Ki-67 was positively correlated with parameters Skewness(r=0.444,p=0.001),Cluster shade(r=0.413,p=0.003),Cluster Prominence(r=0.564,p=0.009),Entropy(r=0.375,p=0.007),Sum Entropy(r=0.371,p=0.007),Haralick Entropy(r=0.375,p=0.007),Inertia(r=0.328,p=0.019),contrast(r=0.346,p=0.013),Difference Variance(r=0.323,p=0.021),Difference Entropy(r=0.350,p=0.012)and Glcm Entropy(r=0.357,p=0.01)and negatively correlated with parameters Energy(r=-0.361,p=0.009),Uniformity(r=-0.366,p=0.008),Correlation(r=-0.337,p=0.016),Angular Second Moment(r=-0.361,p=0.009)and Haralick Inverse Difference Moment(r=-0.371,p=0.007).Conclusion: TA performed on conventional MRI sequences to differentiate cerebral gliomas,taking in account their genetic changes,showed good and promising results which may help and improve the management of these lesions.