Design and Development of Integrated Deep Convolution Neural Network Approach for Handling Heterogeneous Medical Data

Abstract

In recent years, Neural Network (NN) is developed as an optimal technique for the prediction of tasks which include image classification, speech recognition and also useful in biomedical analysis. Biomedical data consists of diverse modalities like X-ray, CT, MRI, PET, EEG and ECG signals. There are several NNs techniques such as Artificial Neural Network (ANN), Convolutional Neural Network (CNN) and Deep Neural Network (DNN) that are used for various prediction applications in handling multimodal heterogeneous data. However, learning and prediction of such multimodal data limits the scope of existing neural techniques. One of the limitations of ANN observes that addition of layer cause back propagation stuck in local minima and reduction of learning speed. Where, DNN causes higher computational complexity in training the features which are based on contrastive divergence. In CNN there is loses of spatial information due to the weight factors variation. To overcome these issues, this paper proposes a novel learning technique in which the weight factor of DNN is integrated with CNN for handling multimodal heterogeneous data. The simulation results prove that the integrated learning technique (IDCNN) obtains better learning performance than ANN, CNN and DNN models in terms of Root Mean Square Error (RMSE) and efficiency in terms of cross entropy.