Classification of Grades of Astrocytoma Images from MRI Using Deep Neural Network

Due to increased size and volume in medical images the automatic diagnosis process is obligatory. Among all other types of cancer, the mortality rate of brain cancer is the highest. The MRI (Magnetic Resonance Imaging) brain images need to be accurately classified into different grades for better treatment decision. Astrocytoma is one of the common type of glioma brain tumors, comprises 34% of brain tumor. In the proposed system, Astrocytoma MRI images is preprocessed using PCNN (Pulse Coupled Neural Network) and median filter. Three class FCM (Fuzzy Clustering Means) thresholding is used for successful segmentation. From the features obtained from DWT (Discrete Wavelet Transform), texture features are extracted using Gray level co-occurrence matrix (GL-CM). Shape attributes are extracted by manipulating Canny edge detection algorithm. Training set consists of feature vectors created from 400 images from two grades of tumor. Classification of 100 test data into different grade is done using classifiers such as Radial Basis Function Neural Network (RBF NN) and Deep Neural Network (DNN).