CMLM/ZhongJing-OMNI
收藏资源简介:
--- license: mit task_categories: - question-answering - text-generation tags: - medical --- # ZhongJing-OMNI: The First Multimodal Benchmark for Evaluating Traditional Chinese Medicine **ZhongJing-OMNI** is the first multimodal benchmark dataset designed to evaluate Traditional Chinese Medicine (TCM) knowledge in large language models. This dataset provides a diverse array of questions and multimodal data, combining visual and textual information to assess the model’s ability to reason through complex TCM diagnostic and therapeutic scenarios. The unique combination of TCM textual knowledge with multimodal tongue diagnosis data sets a new standard for AI research in TCM. ### Key Multimodal Features: - **Multiple-choice questions**: Encompassing core TCM concepts, syndromes, diagnostics, and herbal formulas. - **Open-ended questions**: Focused on detailed diagnostic reasoning, treatment strategies, and explanation of TCM principles. - **Case-based questions**: Real-world clinical cases that require in-depth analysis and comprehensive treatment approaches. - **Multimodal tongue diagnosis Q&A**: High-resolution tongue images paired with corresponding diagnostic questions and expert answers, combining both visual and textual data to evaluate the model’s understanding of TCM tongue diagnosis. This multimodal dataset allows AI systems to develop a deeper, more holistic understanding of TCM by integrating textual reasoning with visual diagnostic skills, making it a powerful resource for healthcare AI research. ## Dataset Structure - `MCQ/`: Multiple-choice questions with answer keys. - `OpenQA/`: Open-ended questions with detailed expert-verified answers. - `CaseQA/`: Clinical case-based questions and answers. - `TongueDiagnosis/`: High-quality tongue diagnosis images with paired Q&A for multimodal analysis. ## How to Use ### 1. Clone the repository: ```bash git clone https://github.com/yourusername/ZhongJing-OMNI.git #2. Load the dataset: ``` ```python import pandas as pd # Load multiple-choice data mcq_data = pd.read_csv('MCQ/questions.csv') # Load open-ended Q&A openqa_data = pd.read_csv('OpenQA/questions.csv') # Load case-based Q&A caseqa_data = pd.read_csv('CaseQA/questions.csv') # Load tongue diagnosis Q&A (multimodal data) tongue_data = pd.read_csv('TongueDiagnosis/tongue_questions.csv') ``` 3. Multimodal Tongue Diagnosis Example: ```python from PIL import Image # Load and display an example tongue image for multimodal evaluation img = Image.open('TongueDiagnosis/images/tongue001.png') img.show() # Load the corresponding Q&A with open('TongueDiagnosis/questions/tongue001_question.txt', 'r') as file: question = file.read() print(f"Question: {question}") with open('TongueDiagnosis/answers/tongue001_answer.txt', 'r') as file: answer = file.read() print(f"Answer: {answer}") ``` ## Why Multimodal? The ZhongJing-OMNI dataset introduces the first multimodal component for TCM, combining visual and textual data, which is crucial for understanding complex diagnostic features such as tongue color, shape, and coating. This allows models to: - **Learn how to integrate visual diagnostic features with textual knowledge. - **Perform joint reasoning over both modalities to reach accurate TCM diagnoses. - **Support real-world clinical applications where visual and textual data are intertwined. # Tongue Diagnosis Example: Qi Deficiency with Pale Tongue  This image shows a pale, slightly swollen tongue with a thin white coating. These features are typical signs of Qi deficiency in Traditional Chinese Medicine. This example represents an actual test result from our dataset using the Claude-3.5-Sonnet model. It demonstrates the model's capability to accurately identify and describe key features of tongue images used in Traditional Chinese Medicine diagnosis. ## Contact For questions or collaboration, please contact at Email: ylkan21@m.fudan.edu.cn Citation If you use ZhongJing-OMNI in your research or project, please cite it as follows: ``` @dataset{zhongjing_omni_2024, title = {ZhongJing-OMNI: The First Multimodal Benchmark for Evaluating Traditional Chinese Medicine}, author = {Kang, Yanlan}, year = {2024}, publisher = {GitHub}, journal = {GitHub repository}, url = {https://github.com/yourusername/ZhongJing-OMNI} } ```
许可证:MIT协议 任务类别: - 问答 - 文本生成 标签: - 医疗 # ZhongJing-OMNI:首个用于评估中医药的多模态基准数据集 **ZhongJing-OMNI** 是首个用于评估大语言模型(Large Language Model, LLM)中医药(Traditional Chinese Medicine, TCM)知识的多模态基准数据集。该数据集涵盖多样化的问题与多模态数据,融合视觉与文本信息,用于评估模型在复杂中医药诊疗场景下的推理能力。中医药文本知识与多模态舌诊数据的独特结合,为中医药领域的人工智能研究树立了全新标准。 ### 关键多模态特性: - **选择题**:涵盖中医药核心概念、证型、诊断方法与方剂。 - **开放式问题**:聚焦于详细的诊断推理、治疗策略与中医药理论阐释。 - **基于病例的问题**:包含真实临床病例,需进行深度分析与综合诊疗方案制定。 - **多模态舌诊问答**:将高分辨率舌部图像与对应的诊断问题及专家解答相结合,融合视觉与文本数据,用于评估模型对中医药舌诊的理解能力。 该多模态数据集通过将文本推理与视觉诊断技能相结合,使人工智能系统能够更深入、全面地理解中医药知识,因此成为医疗人工智能研究的优质资源。 ## 数据集结构 - `MCQ/`:包含参考答案的选择题数据集。 - `OpenQA/`:包含经专家详细校验的解答的开放式问答数据集。 - `CaseQA/`:基于临床病例的问答数据集。 - `TongueDiagnosis/`:用于多模态分析的高质量舌诊图像及配套问答数据集。 ## 使用方法 ### 1. 克隆仓库: bash git clone https://github.com/yourusername/ZhongJing-OMNI.git # 2. 加载数据集: python import pandas as pd # 加载选择题数据 mcq_data = pd.read_csv('MCQ/questions.csv') # 加载开放式问答数据 openqa_data = pd.read_csv('OpenQA/questions.csv') # 加载基于病例的问答数据 caseqa_data = pd.read_csv('CaseQA/questions.csv') # 加载舌诊问答(多模态数据) tongue_data = pd.read_csv('TongueDiagnosis/tongue_questions.csv') 3. 多模态舌诊示例: python from PIL import Image # 加载并展示用于多模态评估的示例舌部图像 img = Image.open('TongueDiagnosis/images/tongue001.png') img.show() # 加载对应的问答数据 with open('TongueDiagnosis/questions/tongue001_question.txt', 'r') as file: question = file.read() print(f"Question: {question}") with open('TongueDiagnosis/answers/tongue001_answer.txt', 'r') as file: answer = file.read() print(f"Answer: {answer}") ## 为何采用多模态? ZhongJing-OMNI数据集引入了首个中医药领域的多模态模块,融合视觉与文本数据,这对于理解舌色、舌形、舌苔等复杂诊断特征至关重要。该数据集可帮助模型实现以下能力: - **掌握如何将视觉诊断特征与文本知识相结合** - **针对两种模态进行联合推理,以得出准确的中医药诊断结果** - **支持视觉与文本数据交织的真实临床应用场景** ## 舌诊示例:气虚证淡舌  该图像展示了淡舌、轻度肿胀伴薄白苔的舌象,这些特征是中医药领域气虚证的典型表现。 本示例为使用Claude-3.5-Sonnet模型在本数据集上的实际测试结果,展示了模型准确识别并描述中医药舌诊关键特征的能力。 ## 联系方式 如有疑问或合作意向,请联系邮箱:ylkan21@m.fudan.edu.cn ## 引用说明 若您在研究或项目中使用ZhongJing-OMNI数据集,请按以下格式引用: @dataset{zhongjing_omni_2024, title = {ZhongJing-OMNI: The First Multimodal Benchmark for Evaluating Traditional Chinese Medicine}, author = {Kang, Yanlan}, year = {2024}, publisher = {GitHub}, journal = {GitHub repository}, url = {https://github.com/yourusername/ZhongJing-OMNI} }




