Diagnosis and Prognosis of Alzheimer?s Disease Via 3D CNN

Mild Cognitive Impairment (MCI) is an early symptom of Alzheimer’s disease (AD). The feature extraction and deep learning architecture of the convolutional neural network in 3D brain images is applied to the problem of Alzheimer’s disease. The Structural Magnetic Resonance (sMRI) and Positron Emission Tomography (PET) image of the patient’s brain are classified according to the vigorousness of the disease and is labelled to be either in MCI or in AD or Normal Control (NC) condition. In this paper, we proposed a model and presented the baseline convolutional CNN with four layers viz., Convolutional layer, Leaky Rectified Linear Unit(LReLU), S3Pool layer and Global average pooling. Further, the 3D image data is used to perform the binary and ternary classifications and its performance are examined. The strength of the network has improved interior resource utilization evaluated with medical images, sMRI and PET on hippocampal ROI. The results of our proposed CNN architecture have achieved an accuracy level of 0.945, 0.859 and 0.748 respectively, when compared to the conventional AlexNet based network. The obtained data from the ADNI database shows better performance with our proposed model.

Super-Resolution of Digital Images using CNN with Leaky ReLU

Image super-resolution (SR) has been used in many real world applications as a preprocessing phase. The improvement in image resolution increases the performance of image analysis process. The SR of digital images is achieved by taking the low resolution images as inputs. In this article, a novel deeplearning based super-resolution approach is proposed. The proposed approach uses Convolutional Neural Network (CNN) with leaky rectified linear unit (ReLU) for learning and generalization. The experiments with test images taken from USC-SIPI dataset indicate that the proposed approach increases the quality of the images in terms of the quantitative metric peak signal to noise ratio.

Maxout Networks for Visual Recognition

This study investigates the effectiveness of multiple maxout activation variants on image classification, facial identification and verification tasks using convolutional neural networks. A network with maxout activation has a higher number of trainable parameters compared to networks with traditional activation functions. However, it is not clear if the activation function itself or the increase in the number of trainable parameters is responsible for yielding the best performance on different entity recognition tasks. This article investigates if an increase in the number of convolutional filters on the rectified linear unit activation performs equal-to or better-than maxout networks. Our experiments compare rectified linear unit, leaky rectified linear unit, scaled exponential linear unit, and hyperbolic tangent to four maxout variants. Throughout the experiments, we found that on average, across all datasets, the rectified linear unit networks perform better than any maxout activation when the number of convolutional filters is increased six times.

Learning Based Resolution Enhancement of Digital Images

Image super-resolution (SR), the process that improves the resolution, has been used in many real world applications. SR is the preprocessing phase of majority of these applications. The improvement in image resolution improves the performance of image analysis process. The SR of digital images take the low resolution images as inputs. In this article, a learning based digital image SR approach is proposed. The proposed approach uses Convolutional Neural Network (CNN) with leaky rectified linear unit (ReLU) for learning and generalization. The experiments with the test dataset from USC-SIPI indicate that the proposed approach increases the quality of the images in terms of the quantitative metric peak signal to noise ratio. Further, it avoided the problem of dying ReLU.