MRI-DWI Radiomics And Transcriptome Expression Characteristics Associated With The Prognosis Ofischemic Stroke Combined With Disease And Syndrome

Objective1.Use radiomics technology to screen radiomics features related to early recurrence and disability prognosis of ischemic stroke,build a radiomics prognosis prediction model,and evaluate the radiomics features based on multicenter datasets in ischemic stroke.The application value in prognosis prediction of disease;2.Construct a combined prediction model combining TCM syndrome elements,clinical risk factors in modern medicine and radiomics,and compare multiple models;3.Explore the miRNA and miRNA-mRNA targeting regulation relationship related to the recurrence of ischemic stroke,reveal the intrinsic biological mechanism related to recurrence,and improve the interpretability of radiomic features by applying the biological functions represented by genes.Methods1.Study on MRI-DWI radiomics characteristics related to ischemic stroke recurrence and disability:the study subjects were patients with acute ischemic stroke,and the main outcome was the recurrence of ischemic stroke patients during the follow-up period,including TIA,cerebral infarction,and cerebral hemorrhage.The secondary outcome measure was the patient’s disability prognosis,which was assessed by mRS score after 3 months(0-2:good prognosis,3-6:poor prognosis).At the time of enrollment,the general information,past history,family history,TCM syndrome,TCM constitution,NIHSS score,mRS score,cranial MR examination,test indicators,etc.of the patients were recorded;the 12th week after enrollment(3 months)the mRS scores of the patients were recorded and followed up for 2 years in combination with telephone calls.Radiomics analysis was performed using DICOM images of cranial MRI-DWI images within 72 hours of onset.The ROI image was segmented by the manual segmentation method,and the radiomics feature extraction was performed on the pyradiomics open source platform.A total of 7 feature types including first order,shape,GLCM,GLRLM,GLSZM,NGTDM,and GLDM were extracted.At the same time,three kinds of filters are used to reduce the noise of the target image.Before feature extraction,the internal parameters are set through "Setting" to realize the resampling and normalization of the original image.Z-Score was used to standardize the data of radiomics feature values,mRMR algorithm was used for dimensionality reduction and screening of omics features and clinical features,and logistic regression model was used to construct a radiomics feature prediction model(rad-score)and a combination of clinical features.The comprehensive model of the radiomics features was visualized using the nomogram model.And use leave-one-out cross-validation,draw ROC curve to evaluate the model.Spearman correlation test was used to analyze the correlation between radiomics signature features and TCM syndrome elements.2.Study on miRNA expression and miRNA-mRNA regulation relationship related to recurrence of ischemic stroke.The research object is the study population of acute ischemic stroke recurrence in the second research content.7 patients with recurrence were randomly selected,and 7 patients without recurrence were matched 1:1 according to their age(±5 years)and gender.There were 14 patients in total.Blood was collected within 7 days of the patient’s onset.The experimental process includes:total RNA extraction and quality control,library construction,on-machine sequencing,sequencing data filtering and quality control,and reference genome comparison.Quantitative analysis of small RNA expression during the research process used specific molecular signatures,DEGseq was used to screen differential genes,and MA-plot-based random sampling model method was used to calculate differential expression.miRanda,RNAhybrid and TargetScan software were used for target gene prediction,and GO database and KEGG database were used for enrichment analysis.According to the Starbase database,the targeting relationship between genes in the regulatory network was identified,and the R software was used to create a visual target network diagram to display the regulatory relationship of miRNA-mRNA.Results1.Study on MRI-DWI radiomic characteristics of ischemic stroke recurrence and disability:(1)A total of 75 patients were collected for recurrence outcome indicators,including 15 patients with recurrence and 60 patients without recurrence.A total of 1037 radiomics features were extracted,and dimensionality reduction was performed on the radiomics features to obtain 7 features with the greatest correlation with recurrence:wavelet.LLH-firstorder_Minimum,wavelet.LHL_glcm_Difference Variance,wavelet.HHL_glcm_MCC,wavelet.HLL_glrlm_Ru nVariance,wavelet.HLL_glcm_Correlation,wavelet.LHL_glcm_InverseVariance,wavelet.LHL_glcm_ClusterProminence.Rad-score(recurrence)=-1.9161527-0.7766219*wavelet.LLH_firstorder_Minimum-0.8060113*wavele t.LHL_glcm_Difference Variance+0.1427865*wavelet.HHL_glcm_MCC-1.0276311*wavelet.HLL_glrlm_RunVariance-0.1635419*wavelet.HLL_glcm_Corre lation-0.7311634*wavelet.HL_glcm_InverseVariance+0.9219086*wavelet.L HL_glcm_ClusterProminence.Include TCM syndrome elements(Qi deficiency syndrome,blood stasis syndrome,phlegm-damp syndrome,Yin deficiency syndrome,fire syndrome,internal wind syndrome),build a prediction model of radiomics combined with TCM syndrome,and obtain the correlation between recurrence of ischemic stroke patients.Rad-score(radiomics combined with TCM syndromes)=-2.84327630-0.81771224*wavelet.LLH_firstorder_Minimum-1.58264990*wavelet.L HL_glcm_DifferenceVariance+0.20585791*wavelet.HHL_glcm_MCC-1.58888131*wavelet.HLL_glrlm_RunVariance-0.02385994*wavelet.HLL_glcm_Correlation-0.73377403*wavelet.LHL_glcm_InverseVariance+1.72894333*wavelet.LHL_glcm_Clust erProminence-18.90938261*internal wind syndrome+1.20700687*blood stasis syndrome+0.3 8518099*qi deficiency syndrome-16.453 5 5707*yin deficiency syndrome+1.05747405*phlegm damp syndrome+1.08698547*fire syndrome.The leave-one-out cross-validation was used to evaluate the radiomics prediction model and the radiomics TCM syndrome combined model,and the average prediction accuracy of a single radiomics rad-score tag was 0.79;The average prediction accuracy of omics TCM syndrome combined with rad-score label is 0.80.The AUC values of the single radiomics prediction model and the joint prediction model were 0.845 and 0.879.(2)A total of 90 patients with poor prognosis were included in the study,including 18 patients with poor prognosis and 72 patients with good prognosis.From 1037 radiomics features,9 features with the greatest correlation with poor prognosis were obtained by dimensionality reduction,including:original_firstorder_Kurtosis,wavelet.LHL_glcm_Correlation,square_glrlm_Ru nVariance,square_firstorder_InterquartileRange,original_shape_MajorAxisLe ngth,wavelet.LHL_glcm_MCC,wavelet.LHL_firstorder Skewness,wavelet.LH L_gldm_DependenceVariance,wavelet.LHL_gldm_Dependence Variance,wavel et.LLL_firstorder_Kurtosis.Construction of a poor prognosis prediction model for ischemic stroke patients based on radiomics features(Rad-score formula):Rad-score(poor prognosis)=-2.2925848882-4.831483475*1 original_firstorder_Kurtosis+0.1927196973*wavelet.LHL_glcm_Correlation+0.0003237453*square_glrlm_RunVariance-0.6117750640*s quare_firstorder_InterquartileRange+0.2798225232*original_shape_MajorAxisLengt h+0.7549377082*wavelet.LHL_glcm_MCC-0.4394640569*wavelet.LHL_firstorder Skewness+1.0825387075*wavelet.LHL_gldm_DependenceVariance+5.5284287144*wavelet.LLL_firstorder_Kurtosis.Combined with clinical characteristics(number of infarct lesions,hypoglycemic drugs,cerebral infarction,TOAST classification,baseline NIHSS score),a combined prediction model of radiomics combined with clinical characteristics of modern medicine was constructed:Rad-score=-23.0324727-4.0221194*original_firstorder_Kurtosis+0.3854199*wavelet.LHL_glcm_Correlation-0.1881209*square_glrlm_RunVariance-0.6713484*square-fi rstorder_InterquartileRange+0.2897716*original_shape_MajorAxisLength+0.6307760*wavelet.LHL_glcm_MCC-0.3904875*wavelet.LHL_firstorder_Skewness+0.9508516*wavelet.LHL_gldm_DependenceVariance+4.7410290*wavelet.LLL_firstorder_K urtosis+0.4773702*number of infarct lesions+0.7502595*hypoglycemic drugs-1.2292947*lobarinfarction+18.6229495*TOAST classification+1.1861231*baseline NIHSS total score.The prediction models related to poor prognosis were evaluated by leave-one-out cross-validation,and the average prediction accuracy of a single radiomics prediction model was 0.84;A combined model of radiomics and modern medical features with an average prediction accuracy of 0.86.The AUC values of the two were 0.88 and 0.922.(3)The interpretability analysis of TCM syndrome elements on the radiomics characteristics showed that the radiomics rad-score values related to internal wind syndrome and recurrence were negatively correlated(r=-0.25,P<0.05),other syndrome elements were not significantly correlated,and there was no statistical significance(P>0.05).2.Study on miRNA expression and miRNA-mRNA regulation relationship related to the recurrence of ischemic stroke:(1)The total RNA of the 14 samples was between 45.0 and 478.2ng/μL,the total RNA was between 0.90 and 19.13 μg,and the average total RNA was 7.24μg.All 14 samples met the conditions for library construction and sequencing.(2)After the original data is filtered,the Q20 of all samples is greater than or equal to 97.4%,and the percentage of tags in the original tags after filtering is greater than or equal to 81.42%.The clean reads obtained after filtering were compared with the reference genome sequence,and the average alignment rate of the sample compared with the genome was 94.4%.According to the gene expression level of each sample,statistical analysis was performed on the detected genes that were significantly differentially expressed between the groups.A total of 30 differentially expressed genes were detected between the recurrence group and the non-recurrence group,including 10 up-regulated genes and 20 down-regulated genes.Target gene prediction of differential miRNAs between groups found that there were 166 differential target genes including TTC22,ZNF362,and ZNF827.(3)Significant GO enrichment analysis shows that target genes are mainly enriched in cellular components such as protein complexes and synapses;Enriched in molecular functions such as protein binding,organic compound binding,cytokine receptor activity,and transcription factor activity;Participate in biological processes such as RNA metabolism,protein localization,and tissue development.The enrichment analysis of Kegg signaling pathway showed that the signaling pathways involved in target genes include:JAK-STAT signaling pathway,FoxO signaling pathway,PI3K-Akt signaling pathway,ErbB signaling pathway,mRNA monitoring pathway,axon regeneration and other pathways.Using miRNA-mRNA network interaction analysis,it was found that six major genes including CDKN1A,CSF3R,and IL7R were involved in the JAK-STAT signaling pathway,hsa-miR-6791-3p,hsa-miR-17-3p,hsa-miR-423-5p and hsa-miR-744-5p are involved in the regulation of this pathway.hsa-miR-17-3p,hsa-miR-4286,hsa-miR-6791-3p,hsa-miR-423-5p,hsa-miR-34a-5p,hsa-miR-497-5p,hsa-miR-744-5p has regulatory relationship with target genes CDKN1A,IL7R,PIK3CD,MAPK13 in FoxO signaling pathway;the main genes CDKN1A,COL9A2,CSF3,IL7R,ITGB5,PDGFRB,PIK3CD and hsa-miR-129-5p,hsa-miR-17-3p,hsa-miR-423-5p,hsa-miR-6791-3p,hsa-miR-744-5p,etc.have targeted interactions in the PI3K-Akt signaling pathway.Relapse-related hsa-miR-6791-3p,hsa-miR-744-5p and genes CDKN1A,COL9A2,CSF3,IL7R,ITGB5,PDGFRB,PIK3CD have targeted interactions in ErbB signaling pathway.Conclusion1.Radiomics prediction models have good predictive value for early recurrence(≤2 years)and poor prognosis(3 months)of ischemic stroke.2.The combined model of TCM syndrome elements and clinical features of modern medicine and radiomics has a beneficial effect on a single radiomics model in terms of predictive performance.3.The recurrence of ischemic stroke may be related to hsa-miR-6791-3p,hsa-miR-17-3p and other genes,and participate in the regulation of JAK-STAT,FoxO,PI3K-Akt,ErbB and other neuronal inflammatory responses,cellular autonomic Phagocytosis and other brain tissue damage signaling pathways.4.The predictive effect of radiomics features on the recurrence of ischemic stroke patients may be related to the neuroinflammation,autophagy and other biological processes of brain injury involving multiple genes.Highlight of innovation1.The first to use a multi-center study design to explore the radiomics feature biomarkers associated with short-term recurrence of ischemic stroke,and to construct and evaluate the application value of a prediction model based on radiomics features for the prediction of short-term recurrence of ischemic stroke.2.Constructed and evaluated the combined prediction model of radiomics and TCM syndrome elements,and innovatively discovered the difference in prediction efficiency between TCM syndrome elements and radiomics features.3.Take the lead in analyzing the interpretability of TCM syndromes based on radiomics features.