Alzheimer_Disease
Description
study uses a synthetic, multimodal benchmark designed to evaluate the proposed method under controlled, reproducible conditions. All inputs are numerically generated; they are analogs of structural neuroimaging, electroencephalography (EEG), and tabular biomarker data, and are used to assess architectural and training behavior, not to claim performance on a specific real-world patient registry. Modalities. Each sample comprises three parts: a single-channel three-dimensional volume (MRI-like) with default spatial layout 1 × 16 × 64 × 64; a multichannel time series (EEG-like) with 19 channels and 256 time points per segment; and a fixed-length biomarker vector of 12 real-valued features. Outcome and class definitions. The learning target is five-class Alzheimer’s disease–related staging, with integer labels 0–4: cognitively normal (CN); subjective memory complaints (SMC); early mild cognitive impairment (EMCI); late mild cognitive impairment (LMCI); and Alzheimer’s disease dementia (AD). Sample construction and size. A stratified reference bundle of 22,000 samples is defined using class proportions [0.26, 0.12, 0.22, 0.22, 0.18] for the five classes, yielding approximate class counts of 5,720, 2,640, 4,840, 4,840, and 3,960. Data are partitioned into training (~80%) and validation (~20%) subsamples with stratification so that class ratios are preserved in expectation. Fixed random seeds are used so that reported metrics and visualizations are reproducible. Generative assumptions. Each modality is produced by class-conditional structure (shared information linked to the label) plus stochastic noise, so that learning is non-trivial but the task remains well posed. The reference setting assumes no missing modalities; extensions to missing or masked inputs are left to future work. Preprocessing. Volumetric inputs are min–max normalized to a bounded intensity range per sample. EEG and biomarker channels may be optionally standardized (e.g., z-score) in extended analyses; the baseline protocol uses the normalization steps above for stable optimization. Limitations and scope. Because the benchmark is synthetic, reported accuracy and related statistics do not establish clinical or regulatory performance on real patients. They should be read as methodological results on a controlled benchmark. Validation on independently acquired, multimodal clinical or research cohorts is outside the present scope and is the natural next step for external generalization and deployment-oriented claims.