AI-Powered Clinical Text Understanding for ICD-10 Code Prediction using Hybrid Embedding and Retrieval Models

A higher level of medical documentation and the need to identify the disease with high precision has made the manual coding of ICD-10 time consuming and subjective exercise. Within the scope of the current paper, the researcher proposes an AI-Based Clinical Text Understanding scheme that will be employed to forecast the ICD-10 code with the help of hybrid embedding and retrieval patterns. The system uses more intricate transformer-based embeddings and a retrieval- augmented generation (RAG) technique, which utilizes FAISS to allow semantic search through a structured ICD-10 knowledge base to be efficient. Similarity search mechanism is used to process clinical notes and embed them with the state-of- the-art language models and map them to the most topical ICD- 10 codes. The hybrid structure can be both highly accurate and scalable with the same time being decipherable by highlighting key phrases that influence predictions. Experimental study on real-world clinical results shows that the method has significant improvements in terms of accuracy, recall and F1-score in comparison to the conventional methods of deep-learning. Such a framework also has the potential of reducing the human error and time wasted on coding, which would be of great solution to healthcare providers and medical coders. Automated clinical text understanding is capable of bringing changes to the healthcare analytics and billing accuracy and patient care with the assistance of the proposed model.