01 / The question
Can a convincing mask still be misleading?
The project combined multilevel thresholding, region growing and mathematical morphology to segment candidate tumour regions in MRI images.
02 / The intuition
A mechanism worth testing.
Inspect both the mask and the evaluation pipeline. A plausible result and an impressive metric can share the same hidden mistake.
03 / The work
Inside the method.
Explore each part of the approach.
01Propose regions
Multi-Otsu thresholding identified initial intensity regions.
02Refine the mask
Seed-based region growing and morphological operations cleaned and extended the candidate mask.
03Examine the score
F1 and overlap measures were used to compare masks with the reference.
04 / The observations
What emerged.
The source notes explicitly flag a perfect F1 result as suspicious and potentially biased. The useful lesson is to audit the metric rather than present that number as an achievement.
This account is based on recorded project notes. Original reports, figures and datasets are not embedded here.
05 / The limits
Where the evidence stops.
The evaluation must be reproduced from original masks and code before any performance claim. Manual seeds and reference construction can strongly affect the result.
The academic foundations