405 / 1241
Validation of segmentation algorithms
is crucial
1. Histopathology (gold standard)
•
Difficult to assess.
•
Changes in volume and orientation may be challenging
2. Validation with phantom measurements
•
Necessary to have
•
Real tumours are often not spherical
•
Cannot substitue validation with pathology
3. Comparison to manual delineations
•
Intra-observer variations
•
Regularized auto-segmentation method may always yield
similar accuracy