Abstract TP62: To Determine the Evolution of Stroke Volume Overtime Using ANFIS

Abstract

Background: There is limited information on evolution of temporal lesion volumetric changes in acute ischemia. Though infarct evolves in natural logarithmic or linear pattern in acute phase [24 hrs] both fail to predict the lesion volume in later phase [48,96,200 hours]. Maximum infarct volume [MIV] measured by these methods don’t reach a plateau at the infinite point in time, not true in clinical medicine. Instead of probability based statistical methods we used Adaptive Neuro-Fuzzy Inference System [ANFIS]. ANFIS utilizes features of human brain fundamental to the learning process, pattern modeling and perceptive inference. The purpose is to identify the best method to predict stroke volume at various time intervals. Methods: We used pooled hemicraniectomy database to select patients with three CT scans. Measurement of the infarct volume [IV] was done using ABC/2 method. Maximum infarct volume [MIV] was calculated using last CT before DHC. Time was measured from stroke onset to CT. Mean measured MIV was used as standard to compare the outcome by various methods. A Pseudo-Monte-Carlo scheme was utilized in improving the accuracy of the classification. Results: There were 66 patients with 3 CT-scans. We used MATLAB-2015 with Fuzzy Logic Toolbox functions to determine the pattern of infarct growth [Linear, Logarithmic, Exponential and ANFIS] and to determine the best predictive method at various time intervals. Of the measured IV on CT 1 and 2, 40 patients [60%] data set was used for training the ANFIS system while remaining 26 [40%] was used for testing. Mean measured MIV was 332±112.21 cm 3 [min 112.86 cm 3 , max 650 cm 3 ] at mean time 67.75±67.39 hour [min 21.30 hours, max 350 hours]. Mean MIV measured by ANFIS 319.05±94.58 cm 3 [p=0.32], Linear 251.77±154.46 cm 3 [p=0.01], Logarithmic 259.68±88.76 cm 3 [p=0.00], and Exponential 157.80±84.05 cm 3 Conclusion: ANFIS predicted the third infarct volume with high degree of accuracy across various time intervals. ANFIS method can provide an excellent model for forecasting infarct volume and growth rate, characterizing the evolution of lesion and pathological stages for mechanistic studies and therapeutic interventions of stroke disease.