Integrating compositional pattern-producing networks and optimized convolution neural networks using deep learning techniques for detecting brain abnormalities

Abstract

In brain abnormality approach, accurate and reliable diagnosis is a critical component, because it provides potential abnormality in tissues as well as functional structures significantly to demarcate surgical plan. In the recent past, various diagnosing procedures have been carried out such as Double Density Discrete Wavelet Transform (DDDWT), Support Vector Machine (SVM) classifier, Convolutional Neural Networks etc., whereas treatment outcome and planning are the critical components. Build upon the successful deep learning approach, a novel brain abnormality diagnostic method has been developed by integrating Compositional Pattern Producing (CPP) and Optimized Convolutional Neural Networks (OCNN) in an unified framework. This hybrid approach includes Independent Component Analysis (ICA) with parallel factor and region of interest (ROI) for preprocessing, training CPP using image patches and fine tuning CPP-OCNN using image slices for segmentation as well as for extraction. This model could segment the images by slices than patches which are more accurate and less complex in segmentation with optimized kernel SVM classifier for abnormality detection. The system is evaluated based on MRI imaging datasets provided by CHB-MIT scalp EEG of micro bleeds dataset and efficiently validated with the help of the experimental results by minimizing the Root mean square and by improving the accuracy, smoothness, correlation, sensitivity and specificity using state of the art techniques.