Text this: Leveraging transformer models to predict cognitive impairment: accuracy, efficiency, and interpretability