Brain Tumor Detection in MRI: Technique and Statistical Validation

Abstract

Two novel fractal-based texture features are exploited for pediatric brain tumor segmentation and classification in MRI. One of the two texture features uses piecewise-triangular-prism-surface-area (PTPSA) algorithm for fractal feature extraction. The other texture feature exploits our novel fractional Brownian motion (fBm) framework that combines both fractal and wavelet analyses for fractal wavelet feature extraction. Three MRI modalities such as Tl (gadolinium-enhanced), T2 and fluid-attenuated inversion-recovery (FLAIR) are exploited in this work. The self-organizing map (SOM) algorithm is used for tumor segmentation. For a total of 204 Tl contrast-enhanced, T2 and FLAIR MR images obtained from nine different pediatric patients, the successful tumor segmentation rate is 100%. Two classification methods, multi-layer feedforward neural network and support vector machine (SVM), are used to classify the tumor regions from non-tumor regions. For neural network classifier, at a threshold value of 0.7, the true positive fraction (TPF) values range from 75% to 100% for different patients, with the average value of 90%. For SVM classifier, the average accuracy rate is 95% and 92% when we use 1/3 and 1/2 of data for testing respectively.