NIMG-14IDENTIFICATION OF TRUE FUNCTIONAL ACTIVITY USING RADIOMIC TEXTURE ANALYSIS OF FUNCTIONAL MRI ECHO PLANNAR IMAGING

Abstract

BACKGROUND: FMRI maps commonly show multiple areas of false positive activity. We aimed to identify areas of true fMRI activation by using complex MRI texture analysis (TA) of raw EPI data, thus enhancing the identification of true activity. METHODS: 12 right handed healthy volunteers were recruited (IRB approved). All scans were performed on a 3.0T GE MR scanner. FMRI consisted of a high resolution 3D T1 weighted structural scan and a gradient echo EPI functional scan. FMRI data analysis was performed using SPM8 software for a sentence completion block design task. T map for each individual was displayed using xjview toolbox and thresholded individually to optimize visualization of language area. Each activity cluster was selected and saved. Board-certified neuroradiologists and neuroscientists classified different clusters into two separate groups Expected (E) and Non Expected (NE) based on their anatomical locations. TA was performed using the mean EPI volume for each individual, and 20 invariant texture features were obtained. Efficacy of TA in fMRI was assessed by decision tree and validated using a 4 fold cross validation. RESULTS: fMRI areas of activity in individual scans were identified based on their texture features with 95.3% accuracy, sensitivity 95.3% and specificity 95.3%. Sum of Variance, Cluster shade, sum of average and information measure of correlation were the most important texture features. These features had higher texture values in the NE ROIs indicating more homogeneous areas. Texture features were able to significantly differentiate areas of true versus false activity. CONCLUSIONS: Extracted features indicated that true activity has more heterogeneous texture compared to non-expected activity Quantitative TA of functional MRI using texture feature can enhance the accuracy and sensitivity of the test to true activity and help eliminate false activation