Cognitive Profiling and Personalized Therapy Recommendation for Dementia through a Language Aware Multi-Model Artificially Intelligent System

Over 55 million people currently have dementia, with an expected 135 million cases by 2050. While 82% of primary care physicians are in the frontlines of dementia care, only 13% get proper training. This paper explores a semi-supervised deep learning approach for dementia profiling and therapy recommendation. Utilizing cosine similarity, BERT, and logistic regression, recall, reasoning, and executive function scores were extracted. A custom neural network produced profiles with an accuracy of 92% while Mini-Batch K-means clustering recommended therapies with a Fowlkes-Mallows Score of 0.91. With a processing time of 4.21 minutes, these findings prove the system’s viability and accessibility.