Apparent diffusion coefficient map-based radiomics model for identifying the ischemic penumbra in acute ischemic stroke
10.3760/cma.j.cn112149-20200506-00654
- VernacularTitle:基于ADC图的影像组学模型在判断急性缺血性脑卒中缺血半暗带的价值
- Author:
Ru ZHANG
;
Zhengqi ZHU
;
Li ZHU
;
Shaofeng DUAN
;
Yaqiong GE
;
Tianle WANG
- From:
Chinese Journal of Radiology
2021;55(4):383-389
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the value of ADC map-based radiomics model for identifying the ischemic penumbra (IP) in acute ischemic stroke (AIS).Methods:From January 2014 to October 2019, data of 241 patients with AIS involving the anterior cerebral circulation within 24 h after stroke onset in the First People′s Hospital of Nantong City was analyzed retrospectively. All patients received routine T 1WI, T 2WI, DWI and dynamic susceptibility contrast-perfusion weighted imaging (DSC-PWI). Considering the PWI-DWI mismatch model as the gold standard for determining IP, patients were divided into the PWI-DWI mismatch (84 cases) and PWI-DWI non-mismatch (157 cases) groups. The ROI of the low signal area and the surrounding area was drawn by two doctors at the maximum level of the lesions on the ADC maps. Then the images were imported into AK analysis software to extract the features. Firstly, the inter-class correlation coefficient was used to screen out the features with high consistency, then the maximum relevance and minimum redundancy (mRMR) and least absolute shrinkage and selection operator (Lasso) regression analysis were used to screen the features. The selected features were used to construct their own radiomics model. ROC curve was used to evaluate the performance of the models, and Delong test was used to compare the area under the curve (AUC) of the two models. Results:After screening, 12 features (LongRunLowGreyLevelEmphasis_angle135_offset7, LongRunLowGreyLevelEmphasis_AllDirection_offset7, GLCMEntropy_AllDirection_offset4_SD, GLCMEnergy_angle45_offset1, ColGE_W11B25_16, ColGE_W11B25_24, HaraEntropy, SurfaceVolumeRatio, Sphericity, Quantile0.025, uniformity and Percentile75) were used to construct the radiomics model based on the low signal area of the ADC map. The area under the ROC curve in the training set was 0.900, and the sensitivity, specificity and accuracy were 84.5%, 81.4% and 83.4%, respectively. The area under the ROC curve in the validation set was 0.870, and the sensitivity, specificity and accuracy were 80.9%, 84.0% and 81.9%, respectively. Eleven features(RunLengthNonuniformity_AllDirection_offset1_SD, ShortRunLowGreyLevelEmphasis_angle45_offset1, HighGreyLevelRunEmphasis_AllDirection_offset1_SD, ShortRunLowGreyLevelEmphasis_AllDirection_offset7, HaralickCorrelation_AllDirection_offset4_SD, ClusterShade_angle45_offset7, InverseDifferenceMoment_AllDirection_offset7_SD, ColGE_W3B20_0, sumAverage, SurfaceVolumeRatio and VolumeMM) were used to construct the radiomics model based on the surrounding area of ADC map. The area under ROC curve in training set was 0.820, the sensitivity, specificity and accuracy were 80.5%, 80.2% and 80.4%, respectively; the area under ROC curve in validation set was 0.800, the sensitivity, specificity and accuracy were 78.7%, 80.0% and 79.2%, respectively. The AUC of the radiomics model based on the low signal area of the ADC map was larger than that based on the surrounding area of the ADC map (training set: Z=3.017, P=0.003; validation set: Z=0.604, P=0.002). Conclusion:The radiomics model based on ADC map has a good diagnostic efficacyin identifying the IP.
