Preliminary application of non-contrast CT radiomics for identification of middle cerebral artery occlusion with negative hyperdense artery sign
Objective To investigate the value of non-contrast CT(NCCT)-based radiomics for identifying acute unilateral middle cerebral artery occlusion(MCAO)with negative hyperdense artery sign(HAS).Methods All 80 patients with acute unilateral MCAO confirmed by angiography(MR angiography[MRA]or CT angiography[CTA]or DSA)and presenting with negative NCCT presentation for HAS were enrolled from January 2015 to June 2023 in the Emergency Department of Stroke Center of Affiliated Hospital of Yangzhou university.On the NCCT images,the occluded segment of the middle cerebral artery on the affected side of each case and the corresponding segment of the vessel on the normal side were used as the regions of interest,and a total of 108 radiomic features were extracted.The least absolute shrinkage and selection operator(LASSO)was used to screen the key features,construct and calculate the radiomics score,and four imaging histology models,support vector machine(SVM),light gradient boosting machine(LightGBM),GradientBoosting and adaptive boosting(AdaBoost),were built respectively to predict MCAO.Predictive performance was evaluated by the area under the receiver operating characteristic curves,and comparisons between the modeled receiver operating characteristic curves were made using the Delong test.Finally,the value of the application of radiological modeling was assessed by clinical decision curve analysis(DCA).Results The NCCT images based on 160 vessels were finally screened for 6 key features,including skewness,energy,gray level size zone matrix(GLSZM)-gray uneven,GLSZM-low gray area emphasis,GLSZM-size area non-uniform standardization,GLSZM-area entropy.The area under the curve(AUC)of the SVM-test was 0.688(95％CI 0.497-0.878)with an accuracy of 0.688;the AUC of the LightGBM-test was 0.787(95％CI 0.620-0.955)with an accuracy of 0.781;the AUC of the GradientBoosting-test was 0.654(95％CI 0.457-0.852)with an accuracy of 0.688;the AUC of the AdaBoost-test was 0.707(95％CI 0.515-0.899)with an accuracy of 0.750.The Delong test showed a statistically significant difference between LightGBM-test and GradientBoosting-test(P=0.040),and no statistically significant difference in performance between the remaining models(all P＞0.05).DCA showed that the LightGBM-test performed better.Conclusion NCCT-based radiomics has good diagnostic efficacy for identifying acute unilateral MCAO with negative HAS,and this conclusion needs to be further verified by multi-center and large sample studies.
国家科技期刊平台
NSTL
万方数据
