Characterization and Classification of Brain Tissue and Stroke Lesions in Non-Contrast Computed Tomography Images of Stroke Patients Using Statistical Texture Descriptors and Artificial Neural Network


 Aim: To characterize and classify stroke lesions normal brain tissue in computed tomography (CT) images using statistical texture descriptors.
 Patients methods: Two experienced radiologists blinded to each other inspected CT of 164 patients identify categorize into ischaemic haemorrhagic subtypes. Four regions interest (ROIs) slice that demonstrated the lesion; two representing lesion were selected. Statistical descriptors namely, co-occurrence matrix, run-length absolute gradient histogram calculated for them. Raw data analysis was performed parameters best discriminate between lesions. Artificial neural network (ANN) used ROIs tissue, radiologists’ identification categorization as gold standard, further analyzed receiver operating characteristic curve.
 Results: Three class discriminated The discriminating matrix sum average namely S1-1 SumAverg, S1-0 SumAverg S0-1 SumAverg. For short run emphasis horizontal, 1350 450 directions features. non-zeros, variance mean. class, mean, 90th 99th percentiles parameters. ANN achieved a sensitivity 0.637, specificity 0.753, false positive rate (FPR) 0.247, negative (FNR) 0.363 with matrix. With 0.544, 0.607, FPR 0.393, FNR 0.456 while 0.546, 0.586, 0.414, 0.454. histogram, 0.947, 0.962, 0.038, 0.053.
 Conclusion: features showed highest classification artificial network.