Use of infarct and penumbra growth rate in Machine Learning Models(MLM) to improve the prediction of patient outcomes like mortality in Acute Ischemic Stroke (AIS) patients immediately post thrombectomy.

Machine Learning (ML) is used to predict Clinical and radiological prognosis indicators like acute ischemic infarct, penumbra volumes, Morbidity, and Mortality, etc. Prognostication immediately post thrombectomy helps planning the treatment follow-up, clinical management, and keep the patients well-informed. We devised an improved machine learning model(MLM) to predict dichotomized 90-day mRS (0-1, >=2) using infarct, penumbra growth rate.

Forty acute ischemic stroke patients underwent Stroke Lesion Segmentation (ISLES data of 2015 and 2017) on Diffusion and Perfusion MRI. Input variables chosen are infarct and penumbra volumes, computed infarct, penumbra growth rates based on volumes and onset time, Thrombolysis in Cerebral Infarction Scale (TICI) score post Thrombectomy.

Infarction and penumbra growth rates calculated over a wide range of volume [11–200 ml] and time [1–5 hours]. First ML model using five Input variables(infarct and penumbra volumes and infarct, penumbra growth rates, TICI score) resulted in F1 score 80% compared to second  ML model using four input variables (time since onset, Infarct and Penumbra Volume, TICI score) resulted in F1 score 75%.The addition of growth rates improved F1 score by 5% in prediction of dichotomized 90-day mRS (0-1, >=2).

The inclusion of infarct, and penumbra growth rates improved dichotomized 90-day mRS (0-1, >=2) in acute ischemic stroke patients immediately post Thrombectomy. This is probably because the infarct and penumbra growth rates consider the growth extent of affected vascular territory rather than statically assessing infarct, penumbra volumes at a single point in time. A well-validated Machine Learning model may have a role in the clinical management of Acute Ischemic stroke.