Medical-research-skills literatureimages-interpretation

Source: https://github.com/aipoch/medical-research-skills

When to Use

• You have a paper converted from PDF to Markdown (including

  ## Page XX

  markers and image links) and need a figure-by-figure interpretation report.
• You need to extract key variables, trends, comparisons, and conclusions primarily from charts/plots rather than from the main text.
• You want to align images in a

  *-images/

  folder with figure numbers (e.g., Figure 1A, Figure 2) using captions and in-text citations.
• You need a standardized, UTF-8 Markdown output suitable for downstream summarization, data extraction, or knowledge base ingestion.
• You must filter out non-figure images (e.g., scanned body text blocks) and interpret only chart-like content.

Key Features

• Parses Markdown image links and locates the corresponding

  *-images/

  directory (e.g.,

  RateSOX2.md

  →

  RateSOX2-images/

  ).
• Opens every image in the images folder without skipping and classifies each as chart / table / schematic / flowchart / body-text block.
• Builds an internal (non-exported) alignment list to map images to figure numbers using captions and body-text citations.
• Interprets only chart-type images (and other figure-like visuals when required), excluding body-text blocks.
• Produces a single structured “Image Interpretation” Markdown report per input file, saved to

  outputs/

  .
• Enforces evidence-based interpretation: rely only on captions, body text, and visible image content; do not speculate.

Dependencies

• pdf-extract

  (version: not specified) — used only when the source is PDF and must be converted to Markdown first.
• Markdown template:

  assets/figure_interpretation_template.md

  (version: not specified)
• Quality/requirements reference:

  references/guide.md

  (version: not specified)

Example Usage

Input layout

skill/
  literatureimages-interpretation/
    inputs/
      RateSOX2.md
    RateSOX2-images/
      image_001.png
      image_002.png
      ...
    assets/
      figure_interpretation_template.md
    references/
      guide.md
    outputs/

Run (conceptual workflow)

1. If starting from PDF, convert to Markdown first:

   pdf-extract RateSOX2.pdf > skill/literatureimages-interpretation/inputs/RateSOX2.md
   

2. Ensure the images folder exists and matches the literature name:

   • inputs/RateSOX2.md

   • RateSOX2-images/

3. Execute the interpretation process:

   • Read

     inputs/RateSOX2.md

     (captions + in-text citations + image links).
   • Open every image in

     RateSOX2-images/

     sequentially.
   • Classify images; keep only chart/figure-like items for interpretation.
   • Align images to figure numbers (e.g., Figure 1A) when possible; otherwise mark as Unassigned.
   • Fill

     assets/figure_interpretation_template.md

     .
   • Write exactly one UTF-8 Markdown output to:

     • outputs/RateSOX2_figure_interpretation.md

       (example name; keep it concise)

Output (must be a single Markdown file)

• Location:

  outputs/

• Content: only the “Image Interpretation” section (do not include the internal image list table)
• Encoding: UTF-8

Implementation Details

• Input assumptions

  • Default input is a PDF-to-Markdown file (

    .md

    ) containing:
    • page markers like

      ## Page XX

    • image links
    • captions and surrounding body text
  • If only a PDF is provided, convert it to Markdown using

    pdf-extract

    before interpretation.

• Image discovery

  • Images are typically stored in a folder named

    *-images/

    matching the literature filename.
  • Use Markdown image links and/or folder naming to locate the correct images.

• Mandatory full pass over images

  • Open every image in the

    *-images/

    folder without skipping.
  • Classify each image into one of:
    • chart/plot
    • table
    • schematic
    • flowchart
    • body text block (to be excluded from interpretation)

• Figure attribution (alignment)

  • Use captions and in-text citations to assign figure identifiers (e.g., Figure 2, Fig. 3B).
  • If attribution cannot be determined, label the item as Unassigned.
  • Maintain an internal alignment list for processing only; do not generate or export any image list file.

• Interpretation scope and constraints

  • Interpret only chart-type (and other figure-like) images that require analysis; exclude body-text blocks.
  • Interpretations must be grounded in:
    • visible content in the image (axes, legends, labels, values, trends)
    • caption text
    • relevant body text citations
  • Do not infer beyond the evidence; if information is missing, write “Not specified”.

• Output rules

  • Use

    assets/figure_interpretation_template.md

    as the structure.
  • Output exactly one Markdown file per input document.
  • Save to

    outputs/

    with a concise filename (avoid redundant phrases).
  • Do not include the internal image list table; output only the final “Image Interpretation” content.
  • Ensure UTF-8 encoding to prevent character corruption.

• Quality checks

  • Follow detailed requirements and checkpoints in

    references/guide.md

    .