Implementing Image Diagnosing and Prediction Framework For Detecting And Classifying Brain MRI Images Using Convolutional Neural Network Architecture

Abstract

This paper aims to provide an automatic diagnosis system for brain MRI images. Brain MRI diagnosis is one of the significant and vital fields in medical data analytics because the brain is the central part that operates the whole human body. It leads to a false diagnosis that degrades the segmentation and classification accuracy. Earlier research works on multiple algorithms for preprocessing, segmentation, feature extraction, and classification, increasing computation time and cost complexity. The medical industry requires a fully automatic, fast, and accurate classifier for brain MRI images due to the rapidly increasing population with more brain-related diseases. Thus, this paper has aimed to use a full learning model to learn and extract all the confidential information of the data to do efficient classification. This paper presents a novel Image Diagnosing and Prediction Framework (IDPF) that involves a Stacked Autoencoder and a Convolutional Neural Network (CNN) model for automatically preprocessing, detecting, and classifying brain MRI images. It uses a training and testing dataset divided from the whole dataset in an 80:20 ratio. The IDPF is evaluated by the testing and validation process, which is verified by any random images taken from the training and test dataset. The efficiency of the CNN is improved by using a stacked autoencoder for preprocessing the input images, which can provide more accurate image data for prediction and classification. The entire process of the paper consists of two stages: preprocessing using a stacked autoencoder for denoising, creating the CNN model by training process, and validating the model by the testing process. Finally,  the results are compared with the existing methods to evaluate the performance.