What you are looking at
Patient and scans: a de-identified melanoma patient from the public autoPET IV Longitudinal-CT research dataset (CC BY-NC 4.0), baseline and follow-up CT. Not a GazeID patient. One of GazeID's development cases, never a sealed test case.
Tumours: outlined by the dataset's radiologists on both scans and linked by them. Volumes and the change shown are from those outlines, not from a GazeID model.
Gone tumours are drawn in their first-scan shape at the place the dataset maps them to in the second scan, so you can see where they were.
Body: bones and organs come from the open TotalSegmentator model run by GazeID on the second scan, turned into meshes by GazeID's pipeline.
What GazeID adds, and where it stands: finding the same tumour again automatically after a tap. On development scans the fixed version re-finds 61% of all tumours and 95–99% of lung tumours when the two scans line up; the sealed re-test is running. Not for clinical use.