1. Alzheimer's disease unveiled: Cutting-edge multi-modal neuroimaging and computational methods for enhanced diagnosis.
- Author
-
Mahmood, Tariq, Rehman, Amjad, Saba, Tanzila, Wang, Yu, and Alamri, Faten S.
- Subjects
MACHINE learning ,DELAYED diagnosis ,ALZHEIMER'S disease ,FEATURE selection ,PARTICLE swarm optimization ,MULTIMODAL user interfaces - Abstract
Alzheimer's disease (AD), characterized by stages like Early and Late Mild Cognitive Impairment (EMCI and LMCI), is a growing global concern. Accurate diagnosis is crucial for delaying its onset. Biomarkers like medical imaging and deep learning technology have been developed to address low early AD diagnosis accuracy. Deep learning algorithms can revolutionize the medical system by creating automatic diagnosis models and promoting artificial intelligence. However, existing studies have limitations, such as not considering complementarity between multiple features and ignoring spatial topological properties in the brain structure. It introduces two novel AD classification methods: Depth Double Deep Learning Method of Linear Attention Network (D3LM-LAN) based on Resnet-50, which incorporates a bilinear residual network structure, dual-channel data input, and hybrid attention module for enhanced PET and MRI multimodal data classification; and a machine learning approach using a Multi-Core Support Vector Machine (MLM-MCSVM) that improves softmax classifiers, optimizing data features and kernel functions via Particle Swarm Optimization. These models aid in effective intervention and personalized treatment planning, while the MLM-MCSVM model optimizes early detection with advanced feature selection. Together, these models enhance understanding of AD's heterogeneity, supporting evidence-based clinical practices and patient care. The study uses MRI and PET data to conduct two- and four-class experiments, testing AD vs. Normal Control (NC), MCI vs. NC, AD vs. MCI, EMCI, and LMCI. D3LM-LAN achieved accuracies of 97.74%, 95.01%, 93.82% and 95.69 % in two-class experiments. MLM-MCSVM further improved classification performance, demonstrating enhanced early diagnosis capabilities with accuracies of 98.59%, 96.53%, 95.77%, and 96.32%. • Proposes a multi-core SVM model for precise four-class AD prediction. • Optimizing metric-constrained feature selection and Kernel function. • Select features and classify AD using multimodal neuroimaging data. • Boosting accuracy and interpretability fine-tuning hyperparameters and attention mechanisms. • Analyzes optimization methods, revealing PSO's efficiency in Alzheimer's disease solution selection. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF