This paper addresses the challenges of surgical visual question answering (VQA) in the context of continual learning (CL), particularly focusing on domain shifts and data imbalances. The authors propose a multimodal large language model (LLM)-assisted multi-teacher CL framework to enhance the performance of surgical VQA tasks. The key contributions include: 1. **Multi-teacher CL Framework**: The framework leverages a multimodal LLM as an additional teacher to bridge knowledge gaps during domain shifts and data imbalances. This approach helps in maintaining strong generalization ability and addressing the limitations of traditional CL methods. 2. **Adaptive Weight Assignment**: An adaptive weight assignment methodology is developed to balance the generalization ability of the LLM and the domain expertise of the old CL model. This ensures that the student model can effectively utilize both sources of knowledge. 3. **New Datasets**: Two new surgical VQA datasets, DAISI-VQA and LRSP-VQA, are constructed to simulate realistic VQA CL scenarios. These datasets introduce significant domain shifts and data imbalances, providing valuable resources for future research. 4. **Experimental Results**: Extensive experiments on four datasets demonstrate the effectiveness of the proposed method, showing superior performance compared to other advanced CL schemes. The method consistently improves accuracy and F-score across all datasets, highlighting its robustness and adaptability. 5. **Discussion and Future Work**: The paper discusses the importance of the LLM's role in the CL process and suggests potential future directions, such as decomposing representations into spatial and temporal spaces and integrating multi-modal data to further enhance performance. Overall, the paper provides a comprehensive solution for surgical VQA in CL settings, offering a valuable reference for future research and practical applications in surgical education.This paper addresses the challenges of surgical visual question answering (VQA) in the context of continual learning (CL), particularly focusing on domain shifts and data imbalances. The authors propose a multimodal large language model (LLM)-assisted multi-teacher CL framework to enhance the performance of surgical VQA tasks. The key contributions include: 1. **Multi-teacher CL Framework**: The framework leverages a multimodal LLM as an additional teacher to bridge knowledge gaps during domain shifts and data imbalances. This approach helps in maintaining strong generalization ability and addressing the limitations of traditional CL methods. 2. **Adaptive Weight Assignment**: An adaptive weight assignment methodology is developed to balance the generalization ability of the LLM and the domain expertise of the old CL model. This ensures that the student model can effectively utilize both sources of knowledge. 3. **New Datasets**: Two new surgical VQA datasets, DAISI-VQA and LRSP-VQA, are constructed to simulate realistic VQA CL scenarios. These datasets introduce significant domain shifts and data imbalances, providing valuable resources for future research. 4. **Experimental Results**: Extensive experiments on four datasets demonstrate the effectiveness of the proposed method, showing superior performance compared to other advanced CL schemes. The method consistently improves accuracy and F-score across all datasets, highlighting its robustness and adaptability. 5. **Discussion and Future Work**: The paper discusses the importance of the LLM's role in the CL process and suggests potential future directions, such as decomposing representations into spatial and temporal spaces and integrating multi-modal data to further enhance performance. Overall, the paper provides a comprehensive solution for surgical VQA in CL settings, offering a valuable reference for future research and practical applications in surgical education.