Create prototype application for identification type of brain tumor based on MRI image using backpropagation neural network
This research aims to propose methods that automatically classify the type of brain tumor. Classification of brain tumor is performed based on the texture feature of the image. The research related to brain tumor image classification is mostly done by extracting the texture features globally so this research segmented the tumor region first before the texture feature extraction process. In this research used 5 types of texture features, namely mean, standard deviation, entropy, homogeneity, contrast. The value obtained from the texture feature extraction is local or only based on the segmentation result of the tumor region then used as input in the classification. Image data is separated into training and testing data using two different methods, namely Holdout and K-fold Cross Validation method. Therefore, classification is done in two stages, namely training and testing. Classification process is done by using Backpropagation Artificial Neural Network.
The measurements of accuracy of six experiments with different parameters obtained an average of testing accuracy using 2 neurons in the hidden layer is 66.65% accuracy with data that has been partitioned using Holdout and 57.18% accuracy with data already partitioned using K-fold Cross Validation, and using 3 neurons in the hidden layer obtain an average of 83.31% accuracy with data that has been partitioned using Holdout and 88.33% with data already partitioned using K-fold Cross Validation.