Natural language processing (NLP) has emerged as a critical technology in healthcare, enabling the analysis of patient feedback to derive actionable insights for improving care quality. NLP techniques focus on understanding emotions and sentiments from diverse data sources such as social media platforms, patient surveys, and reviews. However, its full potential is often hindered by challenges related to processing domain-specific medical language and handling large volumes of data in real time. In this study, we propose a sentiment analysis framework that combines the Optimized TF-IDF model with transformers to enhance accuracy and reduce noise in patient sentiment classification. We employed various methods, including Bag of Words, an improved TF-IDF model, and word embeddings, to evaluate patient sentiments. The proposed model demonstrated superior performance, achieving an accuracy of 94% and an F1-score of 0.92, surpassing traditional sentiment analysis approaches. The optimized TF-IDF model proved especially effective in refining term importance and mitigating irrelevant data, thereby delivering more precise insights into patient experiences. These results suggest that implementing NLP-based sentiment analysis in healthcare can significantly improve service quality by providing deeper insights into patient needs, ultimately enhancing satisfaction and care outcomes.