[The neural network algorithm for diagnosis of ischemic stroke pathogenetic subtypes].

The pathogenetic subtype of ischemic stroke (IS) in 204 cases of acute ischemic stroke was based on the results of clinical estimation, computer tomography, duplex scanning of brain vessels, electrocardiography, echocardiography and Holter monitoring. Initially, experts assigned a diagnosis of large-artery atherosclerosis to 59 (29%) patients, cardio embolic subtype to 51 (25%), small-vessel occlusion to 43 (21%), stroke of other etiology to 10 (5%), stroke of unclear etiology to 41 (20%). The TOAST computerized algorithm has been used for stroke subtypes diagnosis but it failed in 73% of the cases. Basing on artificial neural network, the authors developed the algorithm for determination of stroke subtypes by estimating probabilities for 3 IS subtypes: atherothrombotic, cardio embolic and lacunar. Its efficacy was proved by high diagnostic sensitivity of 97% (95% CI, 93% to 99%); positive predictive value--98% ((95% CI, 94% to 99%) and index of agreement with expert (K=0.955; 95% CI, 0.911 to 0.998).