1 - USE OF ARTIFICIAL INTELLIGENCE TO PREDICT VOLUME OF BRAIN AND INFARCTION GROWTH RATE; A STEP TOWARDS PRECISION MEDICINE

Download or read book 1 - USE OF ARTIFICIAL INTELLIGENCE TO PREDICT VOLUME OF BRAIN AND INFARCTION GROWTH RATE; A STEP TOWARDS PRECISION MEDICINE written by Saadat Kamran and published by . This book was released on 2017 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: Background: Background. The final human brain infarction volume [IV] and infarction growth rate [IGR] are strong predictors of clinical outcome. IGR is dynamic with wide variations in speed of growth. The conventional mathematical techniques are unable to predict IV and IGR.Patients and Methods: In this, single-center, prospective study data was collected on all acute stroke patients treated by intravenous thrombolysis and/or mechanical thrombectomy between January 2014 and December 2016.Infarct Growth Rate calculation [IGR]. Infarct volumes were measured on the baseline and 24-hour CT. For IGR calculation we assumed the stroke volume to be zero prior to stroke onset.Infarct growth rate 1[IGR1] = u0394 volume (IV CT1u20130)/u0394 time (time CT1- stroke onset time)Second infarct growth rate [IGR2] was measured on second CT [CT2]IGR2 = u0394 volume/ u0394 time = (IV CT2- IV CT1)/ (time CT2-time CT1).To quantify the difference between the estimated IGR and actual IGR mean square error [MSE] was usedufffdufffdufffdufffdufffdufffd=1/ufffdufffd u2211128_(ufffdufffd=0)^(ufffdufffdu22121)u2592(u3016ufffdufffdufffdufffdufffdufffdu3017_(u3016ufffdufffdufffdufffdufffdufffdufffdufffdufffdufffdufffdufffdu3017_ufffdufffd )u2212u3016ufffdufffdufffdufffdufffdufffdu3017_(u3016ufffdufffdufffdufffdufffdufffdufffdufffdufffdufffdufffdufffdufffdufffdufffdufffdufffdufffdu3017_ufffdufffd ) )^2ResultsA total of 134 consecutive patients with an acute ischemic infarction secondary to middle cerebral artery occlusion were treated with a mean time to treatment from symptom onset of 213.27 +/- 227.12 minutes. A bivariate analysis showed the clot burden score [p=0.003], and time to treatment [p=0.78] was negatively/inversely correlated with IGR, while IGR 1 positively correlated with IGR2 [p=0.067]. The IGR2 was significantly higher when the collateral circulation score was low compared to a high score [p=0.001]. An unfavourable modified treatment in cerebral infarction (mTICI) score had a significantly higher IGR2 compared to those who had a favourable mTICI score [p=0.035] (Table 2 methods section). The demographics, clinical and radiological details are provided in the methods section [Table 2 methods section]. A comparison of ANFIS training and testing data [Table 3, methods section] showed no statistically significant difference except better collateral score [p=0.024] in the testing group with lower IGR2 [p=0.03]. The ANFIS based model was able to predict the IGR2 and infarction volume, calculated from the predicted IGR 2, without any statistically significant difference compared to the original data [p=0.001] [Figure 1, Table 2]. The mean square error was 8.95% with an accuracy of 91.05%. In addition, ANFIS-predicted values were in agreement with the original data as shown by skewness, cross-correlation and Cosine Similarity [Table 2].Conclusions : We showed an ANFIS based model to predict second brain IV [IV2] and second IGR [IGR2] depending on first imaging study, using prospectively collected data of acute stroke patients. The model predicted the IGR2 and IV2 without any significant difference to the original data [p=0.001]. We achieved an accuracy of 91.05% in predicting IV2 and IGR2 by combining demographic, clinical and radiological data and then applying ANFIS. Our study has the potential to help in more effective patient selection for treatment, particular therapies like extended hours thrombectomy and hemicraniectomy for malignant brain strokes and predict outcome in ischemic stroke, a step towards precision medicine.