Use case · Idea to test
Medical-image landmarks and measurements
Suggest editable points for a specialist.
- Scan and defined target
- Suggest landmarks
- Review the measurement
An image viewer could suggest anatomical landmarks or an initial outline in MRI, CT or ultrasound. A specialist could correct the geometry before using a measurement.
Where the small model fits
A modality-specific model returns points, boxes or masks directly. The viewer maps them into image coordinates and physical units using the scan metadata; it does not need to parse an LLM's written coordinates.
What would need to work
Evaluate localization error, wrong or absent targets, measurement error and correction time on the chosen modality. A screenshot-button locator is not a medical landmark model, and no such model has been tested here.
Research proposal only. SkinDeep has no clinical model, patient-use demo or validated medical performance for this task.