Automated Medical Coding Using a Hybrid Decision Tree with Deep Learning Nodes

Accurate medical coding using the International Classification of Diseases (ICD) is essential for healthcare communication, billing, and research. However, traditional manual coding is both time-consuming and error-prone, requiring specialized expertise to assign codes to patient records, such as doctors' notes. While hierarchical deep learning models have shown promise in automating ICD coding, the large and complex label space presents significant challenges to computational efficiency and scalability. To address these limitations, we introduce a Hybrid Decision Tree (HDT)-based classification framework that integrates rule-based logic and deep learning methods. The HDT approach decomposes the coding task into a hierarchy of manageable subtasks, using statistical feature scoring rules for simpler classifications and deep learning models for more complex cases. Experimental results demonstrate that our hybrid approach is a scalable and efficient solution for automated ICD coding, outperforming both the pure decision tree and the full deep learning-based decision tree approaches, by achieving high accuracy with significantly reduced computational overhead.