Persistent DWI-PWI Mismatch Prediction: A Preliminary Study with Two-dimensional Texture Analysis Based on ADC maps

Abstract

Abstract MR DWI-PWI mismatch has been established as a reliable biomarker for ischemic penumbra in acute ischemic stroke(AIS) patients. Texture analysis has demonstrated its ability to predict DWI-PWI mismatch of AIS within 24 hours. The current study was to investigate whether texture analysis of diffusion ADC maps can predict the existence of the DWI-PWI mismatch in the delayed imaging AIS patients. MR diffusion ADC maps of 82 patients with acute cerebral infarction beyond 24 hours from ischemic stroke onset were analyzed (41 with DWI-PWI mismatch and 41 without DWI-PWI mismatch). 2D Texture features were extracted to develop models for predicting the existence of DWI-PWI mismatch. 30 texture features were selected with the texture analysis software, and reduced to 10 with Lasso method and 3 with stepwise logistic regression. An equation was developed with these three features with acceptable diagnostic performance (sensitivity: 80.5%(65.1%~91.2%, specificity: 75.6%(59.7%~ 87.6%), PPV: 76.7%(61.4%~88.2%), NPV 79.5%( 63.5%~90.7%), AUC: 0.859(0.777~0.941), misclassification rate: 21.95%). During the leave-one-out validation, the equation is relatively stable, with 4 case changed the predicted status. 2D texture analysis of diffusion ADC maps can be used to predict the existence of DWI-PWI mismatch in the delayed imaging population. The logistic regression-based equation is a promising predicting method.