--- license: apache-2.0 task_categories: - question-answering - text-generation language: - zh tags: - medical - biology - chemistry size_categories: - 100K 🌐 Github • 🌐 Website • 🤗 HuggingFace ## 🌈 Update * **[2024.02.21]** The [answers](https://github.com/FreedomIntelligence/CMB/tree/main/data) to the CMB-Exam test has been updated and some errors caused by omissions in version management have been fixed. * **[2024.01.08]** In order to facilitate testing, we disclose the [answers](https://github.com/FreedomIntelligence/CMB/tree/main/data) to the CMB-Exam test * **[2023.09.22]** CMB is included in [OpenCompass](https://github.com/open-compass/opencompass). * **[2023.08.21]** [Paper](https://arxiv.org/abs/2308.08833) released. * **[2023.08.01]** 🎉🎉🎉 CMB is published！🎉🎉🎉 ## 🌐 Download Data - (Recommended) Download the [zip file](https://github.com/FreedomIntelligence/CMB/tree/main/data) and unzip: ```bash git clone "https://github.com/FreedomIntelligence/CMB.git" && cd CMB && unzip "./data/CMB.zip" -d "./data/" && rm "./data/CMB.zip" ``` - Or load our data as follows: ```python from datasets import load_dataset # CMB-Exam datasets （multiple-choice and multiple-answer questions） exam_datasets = load_dataset('FreedomIntelligence/CMB','exam') # CMB-Clin datasets clin_datasets = load_dataset('FreedomIntelligence/CMB','clin') ``` ## 🥇 Leaderboard Please Check [Leaderboard](https://cmedbenchmark.llmzoo.com/static/leaderboard.html). ## 🥸 Dataset intro ![CMB](assets/CMB-2.svg) ### Components - CMB-Exam: Comprehensive multi-level assessment for medical knowledge - Structure: 6 major categories and 28 subcategories, [View Catalog](catalog.md) - CMB-test: 400 questions per subcategories, 11200 questions in total - CMB-val: 280 questions with solutions and explanations; used as source for CoT and few-shot - CMB-train: 269359 questions for medical knowledge injection - CMB-Clin: 74 cases of complex medical inquires ### CMB-Exam Item ```json { "exam_type": "医师考试", "exam_class": "执业医师", "exam_subject": "口腔执业医师", "question": "患者，男性，11岁。近2个月来时有低热（37～38℃），全身无明显症状。查体无明显阳性体征。X线检查发现右肺中部有一直径约0.8cm类圆形病灶，边缘稍模糊，肺门淋巴结肿大。此男孩可能患", "answer": "D", "question_type": "单项选择题", "option": { "A": "小叶型肺炎", "B": "浸润性肺结核", "C": "继发性肺结核", "D": "原发性肺结核", "E": "粟粒型肺结核" } }, ``` - exam_type: major category - exam_class: sub-category - exam_subject: Specific departments or subdivisions of disciplines - question_type: *multiple-choice (单项选择题)* or *multiple-answer (多项选择题)* ### CMB-Clin Item ```json { "id": 0, "title": "案例分析-腹外疝", "description": "现病史\n（1）病史摘要\n 病人，男，49岁，3小时前解大便后出现右下腹疼痛，右下腹可触及一包块，既往体健。\n（2）主诉\n 右下腹痛并自扪及包块3小时。\n\n体格检查\n体温： T 37.8℃，P 101次／分，呼吸22次/分，BP 100/60mmHg，腹软，未见胃肠型蠕动波，肝脾肋下未及，于右侧腹股沟区可扪及一圆形肿块，约4cm×4cm大小，有压痛、界欠清，且肿块位于腹股沟韧带上内方。\n\n辅助检查\n（1）实验室检查\n 血常规：WBC 5.0×109／L，N 78％。\n 尿常规正常。\n（2）多普勒超声检查\n 沿腹股沟纵切可见一多层分布的混合回声区，宽窄不等，远端膨大，边界整齐，长约4～5cm。\n（3）腹部X线检查\n 可见阶梯状液气平。", "QA_pairs": [ { "question": "简述该病人的诊断及诊断依据。", "solution": "诊断：嵌顿性腹股沟斜疝合并肠梗阻。\n诊断依据：\n①右下腹痛并自扪及包块3小时；\n②有腹胀、呕吐，类似肠梗阻表现；腹部平片可见阶梯状液平，考虑肠梗阻可能；腹部B超考虑，\n腹部包块内可能为肠管可能；\n③有轻度毒性反应或是中毒反应，如 T 37.8℃，P 101次／分，白细胞中性分类78％；\n④腹股沟区包块位于腹股沟韧带上内方。" }, { "question": "简述该病人的鉴别诊断。", "solution": "（1）睾丸鞘膜积液：鞘膜积液所呈现的肿块完全局限在阴囊内，其上界可以清楚地摸到；用透光试验检查肿块，鞘膜积液多为透光（阳性），而疝块则不能透光。\n（2）交通性鞘膜积液：肿块的外形与睾丸鞘膜积液相似。于每日起床后或站立活动时肿块缓慢地出现并增大。平卧或睡觉后肿块逐渐缩小，挤压肿块，其体积也可逐渐缩小。透光试验为阳性。\n（3）精索鞘膜积液：肿块较小，在腹股沟管内，牵拉同侧睾丸可见肿块移动。\n（4）隐睾：腹股沟管内下降不全的睾丸可被误诊为斜疝或精索鞘膜积液。隐睾肿块较小，挤压时可出现特有的胀痛感觉。如患侧阴囊内睾丸缺如，则诊断更为明确。\n（5）急性肠梗阻：肠管被嵌顿的疝可伴发急性肠梗阻，但不应仅满足于肠梗阻的诊断而忽略疝的存在；尤其是病人比较肥胖或疝块较小时，更易发生这类问题而导致治疗上的错误。\n（6）此外，腹股沟区肿块还应与以下疾病鉴别:肿大的淋巴结、动（静）脉瘤、软组织肿瘤、脓肿、\n圆韧带囊肿、子宫内膜异位症等。" }, { "question": "简述该病人的治疗原则。", "solution": "嵌顿性疝原则上需要紧急手术治疗，以防止疝内容物坏死并解除伴发的肠梗阻。术前应做好必要的准备，如有脱水和电解质紊乱，应迅速补液加以纠正。手术的关键在于正确判断疝内容物的活力，然后根据病情确定处理方法。在扩张或切开疝环、解除疝环压迫的前提下，凡肠管呈紫黑色，失去光泽和弹性，刺激后无蠕动和相应肠系膜内无动脉搏动者，即可判定为肠坏死。如肠管尚未坏死，则可将其送回腹腔，按一般易复性疝处理，即行疝囊高位结扎+疝修补术。如肠管确已坏死或一时不能肯定肠管是否已失去活力时，则应在病人全身情况允许的前提下，切除该段肠管并进行一期吻合。凡施行肠切除吻合术的病人，因手术区污染，在高位结扎疝囊后，一般不宜作疝修补术，以免因感染而致修补失败。" } ] }, ``` - title: name of disease - description: information of patient - QA_pairs: a series of questions and their solutions based on the description ## ℹ️ How to evaluate and submit refer to [link](https://github.com/FreedomIntelligence/CMB) ## 😘 Citation Please use the following citation if you intend to use our dataset for training or evaluation: ``` @misc{cmedbenchmark, title={CMB: Chinese Medical Benchmark}, author={Xidong Wang*, Guiming Hardy Chen*, Dingjie Song*, Zhiyi Zhang*, Qingying Xiao, Xiangbo Wu, Feng Jiang, Jianquan Li, Benyou Wang}, note={Xidong Wang, Guiming Hardy Chen, Dingjie Song, and Zhiyi Zhang contributed equally to this github repo.}, year = {2023}, publisher = {GitHub}, journal = {GitHub repository}, howpublished = {\url{https://github.com/FreedomIntelligence/CMB}}, } ``` ## Acknowledgement - We thank [Shenzhen Research Institute of Big Data](http://www.sribd.cn/) for their enormous support for this project. - We thank the following doctors for participating in the human evaluation of CMB-Clin: - 林士军 (香港中文大学（深圳）附属第二医院) - 常河 - 许晓爽