|
|
--- |
|
|
license: cc-by-nc-4.0 |
|
|
language: |
|
|
- en |
|
|
tags: |
|
|
- medical |
|
|
- rheumatoid-arthritis |
|
|
- healthcare |
|
|
- diagnosis |
|
|
pretty_name: Pre-screening Rheumatoid Arthritis Information Database |
|
|
size_categories: |
|
|
- n<1K |
|
|
--- |
|
|
|
|
|
# PreRAID Dataset |
|
|
|
|
|
**Prescreening Rheumatoid Arthritis Information Database (PreRAID)** |
|
|
Developed by RespAI Lab at KIIT and KIMS Bhubaneswar. |
|
|
|
|
|
--- |
|
|
|
|
|
## Overview |
|
|
PreRAID is a structured dataset designed to evaluate the diagnostic capabilities of Large Language Models (LLMs) in Rheumatoid Arthritis (RA) diagnosis. This dataset provides real-world patient data, offering insights into RA prediction and reasoning accuracy. |
|
|
|
|
|
--- |
|
|
|
|
|
## Data Description |
|
|
- **Total Records**: 160 patient entries. |
|
|
- **Collection Location**: KIMS Bhubaneswar, India. |
|
|
- **Demographic Breakdown**: |
|
|
- Gender: 85% Female, 15% Male. |
|
|
- Diagnosis: 85% RA, 15% Non-RA. |
|
|
- **Languages Used**: English and Odia. |
|
|
- **Data Collection**: Through a structured online form supervised by RA medical professionals. |
|
|
|
|
|
### Key Information Captured |
|
|
1. **Demographic Details**: Age, gender, language, and unique identifiers (e.g., KIMS ID). |
|
|
2. **Symptoms**: Pain localization, onset duration, joint swelling, stiffness, and deformities. |
|
|
3. **Associated Conditions**: Skin rashes, fever, ocular discomfort, and daily activity impacts. |
|
|
4. **Doctor-Verified Diagnoses**: Ground truth and explanatory notes for RA and non-RA cases. |
|
|
|
|
|
--- |
|
|
|
|
|
## Dataset Features |
|
|
1. **Structured Patient Records**: Standardized text representation for uniform analysis. |
|
|
2. **Visual Aids**: Diagrams for precise pain localization. |
|
|
3. **Embedded Vectors**: Text embeddings for semantic relationships using GPT-4 text embedding models. |
|
|
4. **Storage**: Organized in a vector database to enable retrieval-augmented generation (RAG). |
|
|
|
|
|
--- |
|
|
|
|
|
## Research Insights |
|
|
The dataset was utilized to investigate LLM misalignment in RA diagnosis. Key findings: |
|
|
- LLMs achieved **95% accuracy** in prediction but with **68% flawed reasoning**. |
|
|
- Misalignment between prediction accuracy and reasoning quality emphasizes the need for reliable explanations in clinical applications. |
|
|
|
|
|
--- |
|
|
|
|
|
## Usage |
|
|
The PreRAID dataset is ideal for: |
|
|
1. **Diagnostic Analysis**: Evaluating AI model accuracy and reasoning quality for RA. |
|
|
2. **RAG Applications**: Utilizing vectorized patient records for enhanced model reasoning. |
|
|
3. **Healthcare AI Research**: Studying interpretability and trustworthiness of LLMs in medical settings. |
|
|
|
|
|
--- |
|
|
|
|
|
## Citation |
|
|
Please cite the following paper when using the PreRAID dataset: |
|
|
|
|
|
``` |
|
|
@misc{maharana2025rightpredictionwrongreasoning, |
|
|
title={Right Prediction, Wrong Reasoning: Uncovering LLM Misalignment in RA Disease Diagnosis}, |
|
|
author={Umakanta Maharana and Sarthak Verma and Avarna Agarwal and Prakashini Mruthyunjaya and Dwarikanath Mahapatra and Sakir Ahmed and Murari Mandal}, |
|
|
year={2025}, |
|
|
eprint={2504.06581}, |
|
|
archivePrefix={arXiv}, |
|
|
primaryClass={cs.AI}, |
|
|
url={https://arxiv.org/abs/2504.06581}, |
|
|
} |
|
|
``` |
|
|
--- |
|
|
|
