Meta-learning for Image Processing Based on Case-Based Reasoning - Computational Intelligence in Healthcare: Advanced Methodologies - page 222

After the image has been processed, the segmentation result is evaluated. This
means that the segmentation quality is judged either by an expert or automatically.
Depending on the obtained evaluation, the knowledge containers (case description,
similarity, solution) are modified to ensure a better segmentation result by processing
again the same image. This task is done by the case-base maintenance unit.
The case-base maintenance unit is shown in Fig. 3. Different from a conventional
segmentation process, CBR also includes the evaluation of the segmentation result
and takes it as a feedback to improve the system performance semi-automatically or
automatically. Usually this is an open problem in many segmentation applications.

Learning Similarity

Case Base

Selective Case
Registration

Case
Generalization

Updated
Case Entering

Case
Formation

Case
Refinement

New Case
Entering

Domain Knowledge
Image Segmentation
Image Characteristics

Fig. 3. Case-base maintenance

When the evaluation of the segmentation result is done manually, the expert com-
pares the original image with the labeled image on display. If the expert detects sig-
nificant differences in the two images, the result is tagged as incorrect and the case-
base management will start.
The evaluation procedure can also be done automatically. However, there is no
general procedure available and evaluation can be done automatically only in a do-
main-dependent fashion.

2.2 Case-Based Maintenance and Model Learning

Case-based maintenance is done for several purposes: 1. to enter a new case, when no
similar cases are available in the case-base, 2. to update an existing case by case re-
finement, and 3. to obtain case generalization.