Tolga D. Dittrich, MD
Wouters A, Robben D, Christensen S, Marquering HA, Roos YBWEM, van Oostenbrugge RJ, van Zwam WH, Dippel DWJ, Majoie CBLM, Schonewille WJ, et al. Prediction of Stroke Infarct Growth Rates by Baseline Perfusion Imaging. Stroke. 2021.
For the acute treatment of ischemic stroke with endovascular therapy (EVT), the time between symptom onset and therapy initiation is considered crucial so far. However, the trend has shifted in recent years from rigid time windows to more individualized, advanced imaging-based, patient selection for EVT.
CT-based perfusion imaging (CTP) has gained importance in identifying individuals with potentially salvageable brain tissue. Automated perfusion assessments using specialized software (e.g., RAPID) are frequently employed in clinical practice to calculate mismatch volume. For the analysis, two key parameters are defined: the relative cerebral blood flow (rCBF) below 30% as a reflection of the ischemic core volume and the delay to the maximum of the residue function (Tmax) of more than 6 seconds, which defines critically hypoperfused brain tissue. The final infarct volume often corresponds with the ischemic core volume determined at baseline in cases of successful reperfusion. In patients without reperfusion, the size of the hypoperfused brain tissue can be used to predict the final infarct size. However, accurate prediction of final infarct size, especially as a function of reperfusion status, is not possible using these conventional CTP analyses as they only represent snapshots at the time of examination.
Alternatives to automated mismatch evaluations are deep learning (DL) approaches. Wouters and colleagues have previously developed a DL-based algorithm that allows the prediction of tissue status based on CTP source images. In their current work, the accuracy of the final infarct volume estimate of the DL approach was compared with that of a classical deconvolution/thresholding analysis in patients with acute ischemic stroke due to large vessel occlusions of the anterior circulation. Clinical parameters of the DL model included time from symptom onset to imaging, time to recanalization, modified Thrombolysis in Cerebral Infarction (mTICI) scores, and occlusion persistence on 24-hour follow-up CT angiography.
Briefly and succinctly summarized, the core results of the study are that the DL-based method improved the estimation of final infarct volume compared with “classical” CTP image evaluation. Moreover, the DL model was also able to predict individual infarct growth rates. But what are the potential implications of these findings for clinical practice? This DL approach is an exciting step in the current evolution away from fixed time windows toward a more individualized “tissue clock”-based continuum in acute stroke therapy. Advantages include the ability to predict final infarct volume based on different scenarios, especially for the time to recanalization, and the ability to differentiate between patients with slow and fast infarct core growth rates. In a therapeutic context, improved final infarct volume estimations would be desirable, e.g., supporting the question of whether a patient should be transferred to a central hospital for late EVT or not.