Knee cartilage segmentation using multi purpose interactive approach
Interactive model incorporates expert interpretation and automated segmentation. However, cartilage has complicated structure, indistinctive tissue contrast in magnetic resonance image of knee hardens image review and existing interactive methods are sensitive to various technical problems such as b...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2016
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/78034/1/GanHongSengPFBME2016.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Interactive model incorporates expert interpretation and automated segmentation. However, cartilage has complicated structure, indistinctive tissue contrast in magnetic resonance image of knee hardens image review and existing interactive methods are sensitive to various technical problems such as bi-label segmentation problem, shortcut problem and sensitive to image noise. Moreover, redundancy issue caused by non-cartilage labelling has never been tackled. Therefore, Bi-Bezier Curve Contrast Enhancement is developed to improve visual quality of magnetic resonance image by considering brightness preservation and contrast enhancement control. Then, Multipurpose Interactive Tool is developed to handle users’ interaction through Label Insertion Point approach. Approximate NonCartilage Labelling system is developed to generate computerized non-cartilage label, while preserves cartilage for expert labelling. Both computerized and interactive labels initialize Random Walks based segmentation model. To evaluate contrast enhancement techniques, Measure of Enhancement (EME), Absolute Mean Brightness Error (AMBE) and Feature Similarity Index (FSIM) are used. The results suggest that Bi-Bezier Curve Contrast Enhancement outperforms existing methods in terms of contrast enhancement control (EME = 41.44±1.06), brightness distortion (AMBE = 14.02±1.29) and image quality (FSIM = 0.92±0.02). Besides, implementation of Approximate Non-Cartilage Labelling model has demonstrated significant efficiency improvement in segmenting normal cartilage (61s±8s, P = 3.52 x 10-5) and diseased cartilage (56s±16s, P = 1.4 x 10-4). Finally, the proposed labelling model has high Dice values (Normal: 0.94±0.022, P = 1.03 x 10-9; Abnormal: 0.92±0.051, P = 4.94 x 10-6) and is found to be beneficial to interactive model (+0.12). |
|---|