Standardization and knowledge integration pose significant challenges in domain-specific learning, especially for low-resource languages and specialized tasks. For International Classification of Diseases (ICD) code assignment, while datasets such as MIMIC offer abundant resources for English-language research, similar resources are scarce in non-English clinical settings. This paper presents CLSurCoder, a cross-lingual transfer learning approach that facilitates knowledge transfer across languages by selecting, transforming, and translating publicly available English datasets. This method significantly enhances the automatic coding capabilities of general-purpose large language models (LLMs) for Chinese Surgical Operation Records (CSOR). To validate our approach, we compiled CSOR-ICD, a dataset comprising de-identified records from six clinical departments, each annotated with expert-validated ICD codes. Experimental results show that CLSurCoder effectively improves model performance for ICD coding in low-resource language environments through transfer learning. This demonstrates a promising solution for addressing the challenges faced in multilingual medical natural language processing. We will make CSOR-ICD publicly available at http://icrc.hitsz.edu.cn/xszy/yjzy.htm .