MRI Brain Images Compression and Classification Using Different Classes of Neural Networks

The aim of this paper is to build an automatic system for compression and classification for magnetic resonance imaging brain images. The algorithm segments the images in order to separate regions of medical interest from its background. Only the regions of interest are compressed with a low-ratio scheme, while the rest of the image is compressed with a high-ratio scheme. Based on Convolutional Neural Network (CNN) method for classification and a Probabilistic Neural Network (PNN) for image segmentation, the system has been developed. Experiments were conducted to evaluate the performance of our approach using different optimizers with a huge dataset of MRI brain images. Results confirmed that the Root Mean Square Propagation (RMSprop) optimizer converges faster with a highest accuracy comparing to other optimizers and showed that the proposed preprocessing schema reduced the execution time.