Interpretable Alzheimer's disease diagnosis based on distance-aware rotary attention with multi-scale cross-complementarity
Year: 2026
Type: article
Abstract: Background Alzheimer's disease (AD) is a neurodegenerative disorder characterized by a progressive decline in memory and cognitive function. Existing methods fail to comprehensively capture the multi-level structural variations across different brain regions, making it difficult to provide reliable diagnostic evidence and pathological insights for clinical use. Methods The interpretable algorithm with distance-aware rotary attention and multi-scale cross-complementary (IDRA-MCC) is first proposed to address the limitations in AD diagnosis and improve the interpretability of model discrimination. The study included 604 subjects from the ADNI dataset (207 AD, 195 cognitively normal (CN), 74 progressive mild cognitive impairment (pMCI), and 128 stable mild cognitive impairment (sMCI) and 196 subjects from the OASIS dataset (105 AD and 91 CN). First, standardized preprocessing is applied to reduce inter-subject variability. Then, a dual-scale patch feature modeling was adopted to construct a distance-aware rotary attention (DRA) module that can explicitly encode relative positional dependencies and sequence-level locality, enhancing the perception of key brain regions. Meanwhile, a multi-scale cross-complementary (MCC) module is designed to achieve cross-scale feature interaction and fusion, capturing multi-level pathological patterns. Finally, an adaptive token aggregation (ATA) mechanism is designed to highlight discriminative responses, and anatomical backtracking is achieved through token-weight visualization, thereby improving model interpretability. Results Interpretability analysis shows that the identified high-response regions are highly consistent with brain regions previously associated with AD and MCI progression, including the precuneus in the pMCI vs. sMCI task. Conclusions IDRA-MCC achieves both high performance and interpretability across different tasks on the ADNI and OASIS datasets, showing potential for AD-related progression assessment and future clinical applications.
Source: Frontiers in Aging Neuroscience
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $3,801
OpenAlex ID: W7225824912