364 / 1241
Problem: Ground truth
•
To calibrate a correct contouring method, the knowldedge
on the correct tumor borders is essential, e.g. from:
- Phantom-measurements
Problem: usually homogenous spheres, glass wall, homogenous background
= not representative for tumors
- simulated images
Problem: extremely harmful to produce, proximity to reality depends on assumptions
- image data with histopathology correlation
not many datasets available, all have shortcomings: shrinking, distortion, problem of
coregistration, diffuse infiltration
- tumor size known from other imaging
Problem: reason for second imaging? other problems in size determination
•
possible surrogates:
- comparison with expert contours, ideally consistent in multiple
observers
- visual or mathematical consensus-contour of different methods
16