An Effective Computer-aided diagnosis Technique for Alzheimer’s Disease Classification using U-net-based Deep Learning

Abstract

The diagnosis of Alzheimer’s disease (AD), a common neurodegenerative disease that impairs thinking and memory abilities in older adults and ultimately results in cognitive impairment and dementia, is made possible in large part by computer-aided diagnosis (CAD). The idea has been to use either machine learning models or deep learning models to develop classification techniques for this disease. CAD techniques and mechanisms have emerged to help and facilitate early detection of this disease as a fundamental step in its treatment plan. As part of our approach, we proposed a model that included the following two pre-processing steps: Contrast Limited Adaptive Histogram Equalization (CLAHE) was utilized to enhance image contrast, especially in low-contrast areas. Normalization was then incorporated to ensure reliable training and faster convergence. A Gray-level co-occurrence matrix technique was used to extract seven texture features from the images following pre-processing: contrast, homogeneity, energy, correlation, variance, dissimilarity, and entropy. After that, these characteristics were added to the model output before the last classification layer. The best hybrid framework out of the five models we examined in this paper was utilized to build a convolutional neural network that can be used to identify AD characteristics from magnetic resonance images. As discussed in Section IV of this article, the U-Net model was selected because of its superior performance. The experimental results demonstrate that this technique showed great accuracy in segmentation and classification for each of the five AD Neuroimaging Initiative categories when a specific diagnosis was made. These results are as follows: Overall, the five classes’ final average scores for the four measures were as follows: 94.46% for Accuracy, 94.32% for Precision, 94.49% for Recall, and 94.41% for F1-score.