American journal of neurodegenerative disease · 2026
Optimizing Alzheimer's prediction with dental care: comparing robust deep learning models in health and retirement study of America.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Taheri Hamed, Musaie Farhan, Mohammadzadeh Nazanin, Tavassoli Naini Parsa, Novin Nargol, Salimi Omid, Amini Alireza, Malekpour Hadis, Panahiyan Parsa, Malek Mahta, Fathiazar Alireza, Namadkolahi Reza
Аннотация
BACKGROUND: Previous research has illustrated links between dental care and Alzheimer's disease, the aim of this study was to examine whether dental care variables can enhance Alzheimer's risk prediction using two models including Long Short-Term Memory (LSTM) and Temporal Convolutional Network (TCN) models with health and retirement study data.
METHODS: 9,979 HRS participants without cognitive impairment were analysed in this study (mean age 67 years; and 59.8% female, waves 11-15: 2006/2008-2016). Analysis included 52 predictors including demographic, genetic (APOE ε4), health (including dental visit frequency), and psychosocial domains. The outcomes as cognitive impairment and dementia were defined by Langa-Kabeto-Weir criteria. Class imbalance was addressed by SMOTE and missing data were mean imputed. We standardized and reshaped features into five-wave temporal sequences. Data were split into 3 sets: 70% training, 10% validation, and 20% test. LSTM (single layer, 64 units) and TCN (two dilated convolutional layers, 32 channels) models were trained for up to 50 epochs using binary cross-entropy loss with Adam optimizer, learning-rate reduction on plateau, and early stopping based on validation F1. The accuracy, precision, recall, F1, and AUC-ROC were evaluated via Five-fold stratified cross-validation; the optimization of classification thresholds was done by maximizing F1. McNemar's test compared final predictions.
RESULTS: It was evident that LSTM consistently outperformed TCN as demonstrated by the following results: test accuracy 99.78% vs 79.01%; AUC-ROC 99.95% vs 92.74%; F1-score 99.67% vs 75.27%. Cross-validation consistency (LSTM F1 ~99.6%±0.1% vs TCN ~76.6%±1.5%) and stable validation-test metrics (final validation loss 0.0136 vs 0.3312) showed robust LSTM performance without overfitting. moreover, McNemar's test showed a significant difference (statistic = 2605.36; P<0.001). models with dental care variables showed high sensitivity (~99.8%).
CONCLUSIONS: Enhancement in Alzheimer's disease risk prediction in older adults was evident when incorporating dental care variables in LSTM-based sequential modeling which suggest potential for early detection. However, further studies in diverse populations and assessment of feature importance is necessary because of reliance on self-reported dental care data and the study's HRS-specific sample.
METHODS: 9,979 HRS participants without cognitive impairment were analysed in this study (mean age 67 years; and 59.8% female, waves 11-15: 2006/2008-2016). Analysis included 52 predictors including demographic, genetic (APOE ε4), health (including dental visit frequency), and psychosocial domains. The outcomes as cognitive impairment and dementia were defined by Langa-Kabeto-Weir criteria. Class imbalance was addressed by SMOTE and missing data were mean imputed. We standardized and reshaped features into five-wave temporal sequences. Data were split into 3 sets: 70% training, 10% validation, and 20% test. LSTM (single layer, 64 units) and TCN (two dilated convolutional layers, 32 channels) models were trained for up to 50 epochs using binary cross-entropy loss with Adam optimizer, learning-rate reduction on plateau, and early stopping based on validation F1. The accuracy, precision, recall, F1, and AUC-ROC were evaluated via Five-fold stratified cross-validation; the optimization of classification thresholds was done by maximizing F1. McNemar's test compared final predictions.
RESULTS: It was evident that LSTM consistently outperformed TCN as demonstrated by the following results: test accuracy 99.78% vs 79.01%; AUC-ROC 99.95% vs 92.74%; F1-score 99.67% vs 75.27%. Cross-validation consistency (LSTM F1 ~99.6%±0.1% vs TCN ~76.6%±1.5%) and stable validation-test metrics (final validation loss 0.0136 vs 0.3312) showed robust LSTM performance without overfitting. moreover, McNemar's test showed a significant difference (statistic = 2605.36; P<0.001). models with dental care variables showed high sensitivity (~99.8%).
CONCLUSIONS: Enhancement in Alzheimer's disease risk prediction in older adults was evident when incorporating dental care variables in LSTM-based sequential modeling which suggest potential for early detection. However, further studies in diverse populations and assessment of feature importance is necessary because of reliance on self-reported dental care data and the study's HRS-specific sample.