Image Based Meta Analysis Plan Document

[jdkent]

Here is the scoping of work for Image Based Meta-Analysis

Assumptions:

• IBMA can be it's own project type since people will not directly compare IBMA results with CBMA results (no algorithm exists to do so).

When creating a project, allow user to select either an IBMA or CBMA project.

Search & Curation

• Search: filters can be applied for images (b/v, t, z) (optional/not mandatory)
  • filters are like a pyramid, if you have b/v you can make t, if you have t you can make z, it all boils down to z.
• some results may not be directly attached to papers?
  • do we only want to represent collections that are associated with papers?
• have all the same curation steps (identification/screening/eligibility/included)

Annotate & Extract

• images of the same name will be assigned to the same analysis, otherwise each image gets it's own analysis (we could do LLM based parsing of this as well)
• filtering

Specify & Execute

[jdkent]

Notes from Feb 9, 2026 Meeting

### Inventory question

- **How many images are available by type?**

### Coverage / scope

- **% of total studies** that would meet the criteria for each algorithm -- based on topics

### Search (priority)

- **Treat text / “NeuroVault” as papers** rather than collections
    
    - ONVOC / Semantic Search 
    - **Filter by has collection**
    - Add paper meta-data to Neurosynth Compose to enable better search

### Extraction Stage

Three Main Challenges

- **Algorithm match**
	- Which studies/images are compatible with my algorithm choice?
	    
- **Selection**
	- Which image is relevant for my scientific question?
	- Meta-data makes it hard to know which image is which

### Quality

- **Heuristics / QA — Julio**
- For MVP run in workflow
- Could pre-compute for all images potential issues


Additional features:
- Contact Authors

[jdkent]

5 challenges:

1. indexing neurovault
2. image labeling
3. putting images into analyses
4. selecting images
5. selecting algorithm
6. showing what's available to users based on their images