Minimal Clinical Benchmark for Alzheimer’s Disease Prediction Using Age and MMSE
2
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Alzheimer’s disease (AD) poses a critical global health challenge, yet many diagnostic approaches such as PET and CSF assays remain invasive, expensive, and inaccessible. This study investigated whether simple, universally available measures—Age and Mini-Mental State Examination (MMSE) scores—can provide a robust predictive baseline for AD classification. Using data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), a harmonized baseline cohort of 3,750 subjects was constructed and analyzed with logistic regression. Five-fold cross-validation ensured robust evaluation, with model performance assessed by AUC, precision, recall, F1-score, and accuracy. The logistic regression model achieved consistent results across folds (mean AUC = 0.929, accuracy ≈ 86%), with balanced precision (≈0.70) and recall (≈0.84), yielding a mean F1-score of ≈0.76. These findings demonstrate that Age and MMSE alone achieve discriminative power comparable to more complex multimodal frameworks while maintaining full clinical interpretability. The proposed benchmark provides a reproducible reference for future multimodal research and a practical, low-cost tool for early risk stratification in resource-constrained healthcare settings.
Abstract
Alzheimer’s disease (AD) poses a critical global health challenge, yet many diagnostic approaches such as PET and CSF assays remain invasive, expensive, and inaccessible. This study investigated whether simple, universally available measures—Age and Mini-Mental State Examination (MMSE) scores—can provide a robust predictive baseline for AD classification. Using data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), a harmonized baseline cohort of 3,750 subjects was constructed and analyzed with logistic regression. Five-fold cross-validation ensured robust evaluation, with model performance assessed by AUC, precision, recall, F1-score, and accuracy. The logistic regression model achieved consistent results across folds (mean AUC = 0.929, accuracy ≈ 86%), with balanced precision (≈0.70) and recall (≈0.84), yielding a mean F1-score of ≈0.76. These findings demonstrate that Age and MMSE alone achieve discriminative power comparable to more complex multimodal frameworks while maintaining full clinical interpretability. The proposed benchmark provides a reproducible reference for future multimodal research and a practical, low-cost tool for early risk stratification in resource-constrained healthcare settings.