The Diagnostic Model Based On~1H-MRS And Diffusion Imaging In Common Intracranial Tumors

Objective：The differential diagnosis of common brain tumors(glioma, meningiomaand intracranial metastases) are difficult in clinical work, but the clinicaltreatment and prognosis of different tumors are different. The project intends todevelop a common intracranial tumor intelligent diagnostic software, to improvediagnostic accuracy of intracranial tumors by combining with the advantages ofMRI technology and artificial intelligence technology. In order to get the nodwhich diagnostic procedure will be more simple.Materials and Methods：All samples including Gliomas25cases (10cases in which low-gradeglioma,15cases of high-grade gliomas), meningioma20cases,15cases ofpatients with metastatic tumors and20normal volunteers were selected fromShandong Medical Imaging Research Institute during November2012-November2013. outine examination and1H-MRS and DWI were performed onthese cases before surgery operation.9out of15metastases were provenclinically, and the remaining cases were confirmed by pathology.Routine examination (axial T2WI, T1WI and FLAIR),1H-MRS, DWI wereperformed preoperatively using Siemens SKYRA3.0T superconductive MRmagnetic resonance. The region of interest (ROI) were respectively placed in thetumor parenchyma, peritumoral edema and contralateral normal area, Theirmetabolites were measured ratios and ADC values, record the ROI of NAA/Cr,Cho/Cr, NAA/Cho ratios and ADC values respectively. To calculate the averageof metabolite ratios and ADC values in three different regions of three tumorswith the mean±standard deviation form using SPSS13.0. By using two-sample tTest, each metabolite ratios and ADC values were compared whether there are differences among the three tumor parenchyma, among the three tumor edemaand the different grade gliomas, which P value less than0.05was consideredstatistically significant.Genetic algorithms were used to set the line feature selection andoptimization from spectral data,20features were extracted as input classifieroptimal feature subset; Typical characteristic value for the metabolite ratios andADC values were input a classifier directly. Fisher discrimination method andsupport vector machine (SVM) classifier were used on two kinds of patientsamples for classification. Then evaluate the classification results based on theright of each single classifier and finally confirm the result. In the actualmedical diagnosis procedure, New cases were input the multiple classifiers, theclassification of artificial intelligence is the last result.Results：(1)Three kinds of tumor parenchyma NAA/Cr were significantly different(p<0.05), meningioma and glioma parenchyma, metastatic tumor parenchyma ofCho/r were significantly different (p<0.05), meningiomas and parenchymaglioma, metastatic tumor parenchyma of NAA/cho were significantly different(p <0.01), meningioma and glioma parenchyma, metastatic tumor parenchymaADC values were significantly different (p<0.01), glioma peritumoral edemaand meningioma, metastases peritumoral edema ADC values were significantlydifferent (p<0.05),(Table3). between high and low-grade gliomas peritumoraledema Cho/r, NAA/Cho and ADC values are statistically significant (p<0.05),(Table4); high-grade gliomas and metastases between peritumoral edema NAA/Cr, Cho/Cr, NAA/Cho and ADC values were statistically significant (p<0.01)(Table5).(2)Genetic Algorithms is employed to extract20features from1H-MRS,ADC values are measured from diffused weighed imaging，and the concentration,the relative ratio of Cho/Cr, NAA/C, Cho/NAA are measured. Extractedfeature and Classic eigenvalue were input Fisher classifier or SVM classifierafter Cross-validation， classification results come out. To Assess diagnosticaccuracy, sensitivity, specificity, positive predictive value and negativepredictive value of the diagnostic model.Conclusions：1. In the tumor parenchyma, there are between meningiomas and two other types of tumors, there are obvious differences between high and low-gradegliomas peritumoral edema, there are obvious differences between thehigh-grade gliomas and metastases. A differential diagnosis can be made byusing of1H-MRS metabolite ratios and ADC values of the tumor andperitumoral edema.2. Artificial intelligence diagnostic model based on1H-MRS and DWI candistinguish features from the magnetic resonance image and spectroscopy whichdoctors difficult to find out, which improves diagnostic accuracy and isconducive to the differential diagnosis. In a certain extent, the diagnostic modelreplace the biopsy, In addition, There is a stronger guiding significance to theclinical surgery, and can be used for evaluation of effect of surgery and detectrecurrence. This subject has good clinical value.