Gather-excite attention-driven deep learning model for Alzheimer disease detection from MRIs using adaptive residual-DenseNet strategy.
Authors
Affiliations (4)
Affiliations (4)
- Department of Computer Science and Engineering, Paavai Engineering College, Namakkal, 637018, Tamilnadu, India.
- Department of Computer Science and Engineering, RamraoAdik Institute of Technology, D.Y.Patil deemed to be university, Navi Mumbai, Maharashtra, 400706, India.
- Department of Electronics and Communication Engineering, Panimalar Engineering College, Varadharajapuram, Poonamallee, Chennai, Tamilnadu, 600123, India.
- Department of Computer Science Engineering, Karpagam College of Engineering, Coimbatore, 641032, Tamilnadu, India. Electronic address: [email protected].
Abstract
In worldwide, Alzheimer's Disease (AD) is one of the leading cause of death and requires earlier diagnosis for timely and effective treatment management. Neuroimaging, particularly Magnetic Resonance Imaging (MRI) provides a promising solution to detect the AD at earlier stages. Timely detection of variations in brain is critical for enabling effective interventions and mitigating the progression of AD. However, manual analysis of MRI scans remains challenging and time-consuming due to structural complexity and variations across individuals. Subtle patterns and background noise often remain undetected with conventional methods, thereby limiting diagnostic accuracy. Computer-based techniques improve reliability, but MRI heterogeneity complicates feature extraction and prediction model development. To overcome the issues of conventional models, the research study proposes Gather-Excite Adaptive Residual-Densenet (GE-ARDNet) model for AD detection. Here, the MRI images are collected and progressed through GE-ARDNet that integrates the residual learning with dense connectivity to enhance feature propagation. The Gather-Excite model exploits contextual information using an attention strategy, focusing on discriminative regions associated with AD pathology. Model efficiency is improved by optimizing Residual-Densenet parameters with Renovated Parameter Masterpiece Optimizer (RPMO), ensuring robust feature learning and generalization. Finally, the cross-validation and independent analysis is validated, demonstrating its effectiveness of the proposed model.