Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mined Prompting and Metadata-Guided Generation for Wound Care Visual Question Answering

Bavana Durgapraveen, Sornaraj Sivasankaran, Abhinand Balachandran, Sriram Rajkumar

TL;DR

The paper tackles the inbox burden in async wound care by proposing two complementary AI approaches to generate clinically relevant free-text responses from text and wound images. It combines mined prompting (retrieving similar training examples to guide generation) with metadata-guided generation (predicting structured wound attributes to condition outputs). Empirical results show that domain-specific models with few-shot prompts achieve high lexical alignment, while metadata grounding improves clinical coherence, suggesting a productive synergy between retrieval and structured reasoning. Despite gains, the study highlights the persistent challenges of clinical precision, model uncertainty, and evaluation in medical multimodal reasoning, pointing to future work on richer datasets and uncertainty-aware frameworks. Overall, the work advances practical directions for AI-assisted wound care by integrating retrieval, structured metadata, and robust evaluation.

Abstract

The rapid expansion of asynchronous remote care has intensified provider workload, creating demand for AI systems that can assist clinicians in managing patient queries more efficiently. The MEDIQA-WV 2025 shared task addresses this challenge by focusing on generating free-text responses to wound care queries paired with images. In this work, we present two complementary approaches developed for the English track. The first leverages a mined prompting strategy, where training data is embedded and the top-k most similar examples are retrieved to serve as few-shot demonstrations during generation. The second approach builds on a metadata ablation study, which identified four metadata attributes that consistently enhance response quality. We train classifiers to predict these attributes for test cases and incorporate them into the generation pipeline, dynamically adjusting outputs based on prediction confidence. Experimental results demonstrate that mined prompting improves response relevance, while metadata-guided generation further refines clinical precision. Together, these methods highlight promising directions for developing AI-driven tools that can provide reliable and efficient wound care support.

Mined Prompting and Metadata-Guided Generation for Wound Care Visual Question Answering

TL;DR

The paper tackles the inbox burden in async wound care by proposing two complementary AI approaches to generate clinically relevant free-text responses from text and wound images. It combines mined prompting (retrieving similar training examples to guide generation) with metadata-guided generation (predicting structured wound attributes to condition outputs). Empirical results show that domain-specific models with few-shot prompts achieve high lexical alignment, while metadata grounding improves clinical coherence, suggesting a productive synergy between retrieval and structured reasoning. Despite gains, the study highlights the persistent challenges of clinical precision, model uncertainty, and evaluation in medical multimodal reasoning, pointing to future work on richer datasets and uncertainty-aware frameworks. Overall, the work advances practical directions for AI-assisted wound care by integrating retrieval, structured metadata, and robust evaluation.

Abstract

The rapid expansion of asynchronous remote care has intensified provider workload, creating demand for AI systems that can assist clinicians in managing patient queries more efficiently. The MEDIQA-WV 2025 shared task addresses this challenge by focusing on generating free-text responses to wound care queries paired with images. In this work, we present two complementary approaches developed for the English track. The first leverages a mined prompting strategy, where training data is embedded and the top-k most similar examples are retrieved to serve as few-shot demonstrations during generation. The second approach builds on a metadata ablation study, which identified four metadata attributes that consistently enhance response quality. We train classifiers to predict these attributes for test cases and incorporate them into the generation pipeline, dynamically adjusting outputs based on prediction confidence. Experimental results demonstrate that mined prompting improves response relevance, while metadata-guided generation further refines clinical precision. Together, these methods highlight promising directions for developing AI-driven tools that can provide reliable and efficient wound care support.

Paper Structure

This paper contains 17 sections, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Shared Task and Dataset
  3. Related Works
  4. Methodology
  5. Model Description
  6. Approach 1: Metadata ablation study and conditional metadata prediction for response generation
  7. Approach 2: Prompt Mining
  8. Evaluation Metrics
  9. Results and Discussion
  10. Performance Comparison
  11. Metadata Ablation Study
  12. Inter-Annotator Agreement and Feature Selection
  13. Comparative Analysis and Clinical Implications
  14. Prompts used in Approach
  15. LLM as Judge Prompt
  16. ...and 2 more sections

Figures (2)

  • Figure 1: The diagram illustrates a workflow where test data (images and queries) are processed by a model to predict metadata, validated by confidence scoring, and transformed into formatted output responses.
  • Figure 2: The Diagram illustrates overview of the mined few-shot prompting strategy with semantic similarity-based example retrieval and model-specific context optimization.