Aphasia, caused by brain damage, affects comprehension, expression, reading, and writing. Early intervention improves outcomes. This study developed a system to help senior aphasia patients practice writing by comparing their writing to correct forms and providing corrections. Using Information and Communication Technology (ICT) and Artificial Intelligence (AI), the system aims to improve re-habilitation efficiency. It uses mobile devices like smartphones and tablets to de-liver personalized handwriting training modules designed for seniors, with terms selected by speech therapists. To enhance text recognition accuracy, the study optimizes the training model using loss functions, focusing on minimizing Cross-Entropy loss. Lower Cross-Entropy values indicate more accurate classification, while values near 1 suggest prediction errors. By averaging these values, the study assesses overfitting and training effectiveness, guiding adjustments in datasets. Cross-validation is used to prevent overfitting and enhance model robust-ness by splitting data into multiple subsets and validating each iteratively. This process refines model parameters and improves accuracy. Experimental results showed that the Inception Net model demonstrated high efficiency and stability, while the Inception Net model had superior accuracy. These enhancements effectively aid in patients’ language recovery.