Source-linked AI summary
MedDialog: Two Large-scale Medical Dialogue Datasets
Xuehai He, Shu Chen, Zeqian Ju, Xiangyu Dong, Hongchao Fang, Sicheng Wang, Yue Yang, Jiaqi Zeng, Ruisi Zhang, Ruoyu Zhang, Meng Zhou, Penghui Zhu, Pengtao Xie
TL;DR
Medical dialogue research needs large, broad conversation collections because existing datasets are difficult to obtain and limited in size or disease coverage. The paper builds two publicly available datasets, MedDialog-EN and MedDialog-CN, and reports that they are the largest English and Chinese medical dialogue datasets, respectively, with broad specialty coverage and diverse patients.
Problem
Existing medical dialogue datasets are difficult to obtain because of privacy concerns and are limited in size or biased toward certain diseases.
Method
The paper builds two large-scale medical dialogue datasets from patient–doctor consultations: MedDialog-EN in English and MedDialog-CN in Chinese.
Results
MedDialog-EN contains about 0.3 million conversations and 0.5 million utterances, while MedDialog-CN contains about 1.1 million conversations and 4 million utterances.
Takeaways & Limitations
The datasets are publicly available, cover broad medical specialties, and include diverse patient populations for medical dialogue research.
Abstract
from arXiv · showhide
Medical dialogue systems are promising in assisting in telemedicine to increase access to healthcare services, improve the quality of patient care, and reduce medical costs. To facilitate the research and development of medical dialogue systems, we build two large-scale medical dialogue datasets: MedDialog-EN and MedDialog-CN. MedDialog-EN is an English dataset containing 0.3 million conversations between patients and doctors and 0.5 million utterances. MedDialog-CN is an Chinese dataset containing 1.1 million conversations and 4 million utterances. To our best knowledge, MedDialog-(EN,CN) are the largest medical dialogue datasets to date. The dataset is available at https://github.com/UCSD-AI4H/Medical-Dialogue-System
1. Introduction
Telemedicine can expand access, lower costs, and improve care quality, but remote care also burdens physicians and makes patient monitoring difficult. Medical dialogue systems and larger, broader datasets are proposed to address these challenges.
- Telemedicine expands access for underserved communities, reduces healthcare costs, and can improve quality of care.Reported benefits include faster, cheaper care, savings exceeding $1,500 per emergency-department visit, and 38% fewer hospital admissions.
- Remote consultations add to physicians’ existing workload and increase the risk of physician burnout.
- Remote patients are difficult for clinicians to track and monitor compared with in-hospital patients.
- Medical dialogue systems aim to act as virtual doctors by asking about patient conditions, providing advice, and proactively checking disease progression.
- Existing medical dialogue datasets are difficult to obtain because of privacy concerns and remain limited in size or biased toward certain diseases.
- MedDialog-EN and MedDialog-CN address these limitations with large, public datasets spanning almost all medical specialties and a wide range of diseases.
2. Datasets
The paper constructs English and Chinese consultation datasets from online healthcare platforms, combining patient-condition descriptions with doctor–patient dialogue and, for Chinese consultations, optional treatment suggestions.
- MedDialog-EN: MedDialog-EN contains 257,454 consultations and 514,908 utterances, with each consultation comprising a condition description and doctor–patient conversation.The data was crawled from iclinic.com and healthcaremagic.com.
- MedDialog-EN: English consultations cover 96 specialties and were conducted from 2008 to 2020.
- MedDialog-CN: MedDialog-CN contains 1,145,231 consultations and 3,959,333 utterances, including condition and history descriptions, doctor–patient conversation, and optional diagnosis or treatment suggestions.
- MedDialog-CN: After merging consecutive utterances from the same speaker, MedDialog-CN contains 3,209,660 utterances, including 1,981,844 from doctors and 1,227,816 from patients.
- MedDialog-CN: Chinese consultations cover 29 broad and 172 fine-grained specialties and were conducted from 2010 to 2020.
3. Advantages of our datasets
MedDialog-EN and MedDialog-CN are presented as the largest English and Chinese medical dialogue datasets, respectively, with scale, specialty coverage, and patient diversity as key advantages.
- The authors describe MedDialog-EN and MedDialog-CN as the largest English and Chinese medical dialogue datasets, respectively.
- Large number of conversations and utterances: MedDialog-EN has about 0.3 million conversations and 0.5 million utterances, while MedDialog-CN has about 1.1 million conversations and 4 million utterances.
- Broad coverage of medical specialities: MedDialog-EN covers about 96 specialty categories, while MedDialog-CN covers about 29 broad and 172 fine-grained specialties.
- Diversity of the patients: Patients in MedDialog-EN come from around the world, and MedDialog-CN includes patients from 31 provincial-level administrative divisions in China.The paper states that this diversity greatly minimizes population biases in the two datasets.
4. Related Works
The related-work comparison reports that the paper’s datasets exceed other medical dialogue datasets in both the number of dialogs and the number of diseases.
- The paper reports that its datasets contain more dialogs than the other datasets shown.
- The paper reports that its datasets contain more diseases than the other datasets shown.
- Table 1 compares the paper’s datasets with several other medical dialogue datasets.
5. Conclusions
The paper builds two large-scale medical dialogue datasets to facilitate research and development in medical dialogue systems. The datasets contain millions of consultations and utterances.
- The datasets are intended to facilitate research and development of medical dialogue systems.
- MedDialog-EN contains 0.3 million conversations and 0.5 million utterances, while MedDialog-CN contains 1.1 million conversations and 4 million utterances.
- The datasets are publicly available and continuously growing.
Description of medical conditions and history
The exemplar consultation records the patient's symptoms and history, followed by a doctor-patient exchange and diagnostic and treatment recommendations.
- The patient reports red, itchy eyes, occasional ulceration, scratching, and recurrence after ointment use.
- The consultation includes a dialogue in which the doctor identifies local eye inflammation and recommends eye ointment twice daily.
- The exemplar consultation is organized into patient history, doctor-patient dialogue, and diagnosis and treatment suggestions.