Iranian Journal of Biomedical Engineering (IJBME)
نوع مقاله : مقاله کامل پژوهشی
نویسندگان
1 دانشجوی دکتری، گروه مهندسی پزشکی، دانشکده مهندسی برق، دانشگاه صنعتی خواجه نصیرالدین طوسی، تهران، ایران
2 گروه مکاترونیک، دانشکده مهندسی برق، دانشگاه صنعتی خواجه نصیرالدین طوسی
3 دانشیار، گروه مغز و اعصاب، دانشکده پزشکی، دانشگاه شهید بهشتی، تهران، ایران
کلیدواژهها
عنوان مقاله English
نویسندگان English
Managing Parkinson’s disease (PD) through medication can be challenging due to variations in symptoms and disease duration. This study aims to demonstrate the potential of sequence-to-sequence algorithms to suggest personalized medication combinations for patients with PD based on their prior visits. The proposed method employs a multi-layer time-aware gated recurrent unit architecture with a multi-head mechanism to accurately predict the dosage of key PD medications based on patients’ motor symptoms and previously prescribed medications. The time-aware model incorporates a decay factor so that older information from more distant visits has less influence on the current state. Our evaluation demonstrates that the proposed architecture achieves high performance in predicting dosages of main PD medication groups with a mean squared error (MSE) of 0.005, a mean absolute error (MAE) of 0.037, a root mean square error (RMSE) of 0.073, and a coefficient of determination (R²) of 0.75. The proposed architecture provides a foundation for developing intelligent clinical decision support systems and personalized medication management tools, offering a data-driven approach to optimizing treatment regimens for patients with PD.
کلیدواژهها English
