ABSTRACT
In this work, it is proposed a method for supervised characterization and classification of tumoral lesions in brain, based on the analysis of irregularities at the lesion contour on T2-weighted MR images. After the choice of a specific image, a segmentation procedure with a threshold selected from the histogram of intensity levels is applied to isolate the lesion, the contour is detected through the application of a gradient operator followed by a conversion to a "time series" using a chain code procedure. The correlation dimension is calculated and analyzed to discriminate between normal or malignant structures. The results found showed that it is possible to detect a differentiation between benign (cysts) and malignant (gliomas) lesions suggesting the potential of this method as a diagnostic tool.
