Research on Large-Language-Model-Based Algorithms for Assisted Diagnosis and Treatment of Anorectal Diseases

Anorectal diseases are prevalent chronic conditions, and traditional Chinese medicine (TCM) diagnosis and treatment often depend on the extensive experience and deep expertise of senior practitioners. However, current large language models (LLMs) face challenges such as static knowledge and inconsistent reasoning, which limit their effectiveness in medical question answering and clinical decision support. Knowledge graphs, as structured representations of medical knowledge, can enrich the static knowledge base of LLMs and enhance their adaptability to clinical diagnostic and therapeutic tasks. In recent years, Chain-of-Thought (CoT) reasoning has demonstrated strong interpretability and controllability in complex reasoning processes, while knowledge-graph-driven Chain-of-Thought reasoning (Graph-CoT) has further improved multi-step reasoning in medical question answering. Building on these advances, this study proposes a fine-tuning approach for LLMs that integrates Knowledge Graphs (KG) with Graph-enhanced Chain-of-Thought (Graph-CoT) reasoning, aiming to improve the accuracy, reliability, and clinical applicability of LLMs in assisting the diagnosis and treatment of anorectal diseases.