AHANet: Adaptive Hybrid Attention Network for Alzheimer’s Disease Classification Using Brain Magnetic Resonance Imaging

Abstract

Alzheimer’s disease (AD) is a progressive neurological problem that causes brain atrophy and affects the memory and thinking skills of an individual. Accurate detection of AD has been a challenging research topic for a long time in the area of medical image processing. Detecting AD at its earliest stage is crucial for the successful treatment of the disease. The proposed Adaptive Hybrid Attention Network (AHANet) has two attention modules, namely Enhanced Non-Local Attention (ENLA) and Coordinate Attention. These modules extract global-level features and local-level features separately from the brain Magnetic Resonance Imaging (MRI), thereby boosting the feature extraction power of the network. The ENLA module extracts spatial and contextual information on a global scale while also capturing important long-range dependencies. The Coordinate Attention module captures local features from the input images. It embeds positional information into the channel attention mechanism for enhanced feature extraction. Moreover, an Adaptive Feature Aggregation (AFA) module is proposed to fuse features from the global and local levels in an effective way. As a result of incorporating the above architectural enhancements into the DenseNet architecture, the proposed network exhibited better performance compared to the existing works. The proposed network was trained and tested on the ADNI dataset, yielding a classification accuracy of 98.53%.

A Dimension Centric Proximate Attention Network and Swin Transformer for Age-Based Classification of Mild Cognitive Impairment From Brain MRI

“…However, it is essential to acknowledge that deep learning is not without its challenges. In the classification of MCI, it is important to consider that subtle or less visually prominent features may possess significant discriminative information for distinguishing between pMCI and sMCI cases [40]. Additionally, the performance of attention models can be hindered if the training dataset does not adequately capture the full diversity and variability of pMCI and sMCI cases.…”

Identifying underlying patterns in Alzheimer's disease trajectory: a deep learning approach and Mendelian randomization analysis

“… 3 , 4 , 5 However, these works make a hard assumption that the rate of deterioration was homogeneous, either between subjects or within different stages of a certain subject, which is a fundamental problem. 6 To address the heterogeneity of AD, various deep learning methods, have been developed to try to reveal the underlying characteristics and patterns of AD 7 , 8 , 9 , 10 , 11 , 12 , 13 or provide time-to-conversion prediction for the progression of AD, 14 , 15 , 16 offering a great potential for early detection and prognosis of AD.…”

Enhancing Alzheimer’s disease diagnosis and staging: a multistage CNN framework using MRI

“…The attention models focus on the most informative image regions. By combining two distinct attention modules (i.e., enhanced non-local attention and coordinate attention), Illakiya et al (32) presented an adaptive hybrid attention network to enhance the performance of the DenseNet architecture, resulting in a higher classification accuracy of 98.53%. Similarly, in another study (33), an integrated model consisting of a depthwise group shuffle, global context network, hybrid multi-focus attention block, and EfficientNEt-B0 was developed to improve the prediction performance of MCI classification.…”