Computer image analysis of brain CT images for discriminating hypodense cerebral lesions in children

A computer software system was designed for the automatic discrimination of focal oedemas from local glioses in brain CT examinations. Image analysis methods were applied to the images of 77 CT examinations of children with focal oedemas (42) or local glioses (35). Textural features derived from the co-occurrence matrix of the lesion's image and a neural network classifier (the multilayer perceptron) were employed for the design of the system. Best classification accuracy (89.6%) was achieved by two textural features (contrast-difference entropy), one hidden layer and three hidden nodes of the classifier. The proposed software system provides new textural information and may be of value to the radiologist in differentiating focal oedemas from local glioses, especially in small lesions, where other radiological criteria are not evident.