Automatic Multimodal Brain Image Classification Using MLP and 3D Glioma Tumor Reconstruction

This research study discusses an enhanced technique for multimodal brain image classification using Multilayer Perceptron (MLP) and introduces tumor location identification and tumor volume measurement techniques. Brain tumor classification and segmentation is an important task in medical image processing. In the proposed method, brain MR image features are extracted by using discrete wavelet transform (DWT) along with absolute Gaussian smooth filters. Supervised binary classification has been used to separate tumorous and non-tumorous images by MLP. The tumor part is segmented from MR images by employing anisotropic diffusion filters (AD F). The boundaries of all segmented tumors are used for volume measurement and 3D reconstruction of the tumor. Based on the 3D tumor model, location of the tumor inside brain is calculated which can help the radiologists in decision making. The proposed technique has been tested on MICCAI BraTS 2015 data. Results show an accuracy of 92.59% in classification of MR images and 90.12% in tumor segmentation and its volume measurements.