| Abstrakt: |
With continuous evolution of deep learning and medical imaging technologies, the prospects for diagnosing and predicting Alzheimer's disease (AD) are becoming brighter. Researchers have found a way to capture the profound characteristics of medical images using convolutional neural networks (CNN). While the Principal Component Analysis Network (PCANet) is a light-weight, deep learning network that creates multi-level filter banks to learn from samples, its adaptability has its limits. As a result, researchers have shifted to non-negative matrix factorization (NMF) to establish a higher order tensor that undergoes tensor decomposition (TD) to reduce the dimensionality in data and generate the end image features. Although deep learning is useful in detecting different stages of the disease using MRI images, there are challenges due to the limited MRI data available, which can make it difficult to achieve the desired level of accuracy. To counter this issue, researchers have developed a Deep Convolutional Generative Adversarial Network (DCGAN) to produce phoney images with features from the input dataset, expanding the number of images available for training and testing. The Generative Adversarial Network (GAN) approach is employed to produce convincing outputs based on the original information. With the use of this simulation method for supervised learning issues, GANs can be trained to learn and identify patterns in data inputs. In order to evaluate the efficacy of the technique, the researchers used three different datasets: two different versions of ADNI and one different version of OASIS. This project is aimed at achieving superior accuracy using minimal computing power, underscoring the importance of scientific progress in our collective pursuit of improved Alzheimer's disease detection and prognosis. [ABSTRACT FROM AUTHOR] |
