[ENH] LLM based annotations on images based on QC reports / IQMs

[nikhil153]

Is there an existing issue for this?

• I have searched the existing issues

New feature

Needs more brainstorming on input, context, and expected outcomes.

Possible use-case(s)

• Provide text description of image quality based on MRIQC IQM reports.

Unclear documentation

No response

[laurapiro17]

i've been thinking about this one. the approach that makes most sense to me is feeding the model a structured summary of percentile-ranked IQMs rather than anything image-derived — something like "CJV: 0.82 (96th percentile vs ABIDE reference)" gives it actual context to reason about instead of a raw number it can't interpret. from a clinical side, what i'd want from the annotation is not just a flag but a brief explanation of which metric is driving it and why it matters for that specific scan type. happy to help prototype some prompt structures if that's useful