Limits...
New region-scalable discriminant and fitting energy functional for driving geometric active contours in medical image segmentation.

Wang X, Niu Y, Tan L, Zhang SX - Comput Math Methods Med (2014)

Bottom Line: The region-scalable discriminant and fitting energy functional is defined to capture the image intensity characteristics in local and global regions for driving the evolution of active contour.The discriminant term in the model aims at separating background and foreground in scalable regions while the fitting term tends to fit the intensity in these regions.The new model utilizes intensity information in the local and global regions as much as possible; so it not only handles better intensity inhomogeneity, but also allows more robustness to noise and more flexible initialization in comparison to the original global region and regional-scalable based models.

View Article: PubMed Central - PubMed

Affiliation: Key Laboratory of Optoelectronic Technology and Systems of Ministry of Education, College of Optoelectronic Engineering, Chongqing University, Chongqing 400044, China.

ABSTRACT
We propose a novel region-based geometric active contour model that uses region-scalable discriminant and fitting energy functional for handling the intensity inhomogeneity and weak boundary problems in medical image segmentation. The region-scalable discriminant and fitting energy functional is defined to capture the image intensity characteristics in local and global regions for driving the evolution of active contour. The discriminant term in the model aims at separating background and foreground in scalable regions while the fitting term tends to fit the intensity in these regions. This model is then transformed into a variational level set formulation with a level set regularization term for accurate computation. The new model utilizes intensity information in the local and global regions as much as possible; so it not only handles better intensity inhomogeneity, but also allows more robustness to noise and more flexible initialization in comparison to the original global region and regional-scalable based models. Experimental results for synthetic and real medical image segmentation show the advantages of the proposed method in terms of accuracy and robustness.

Show MeSH

Related in: MedlinePlus

Testing images in quantitative experiments.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4109108&req=5

fig7: Testing images in quantitative experiments.

Mentions: To quantitatively evaluate the performance of our proposed model, we use it to segment five brain MRI images of ground-truth object (i.e., white matter) and compare it to CV, RSF, and LGIF methods. The 2D MR images are taken from McGill Brain Web [31] with 3% noise level and 40% intensity nonuniformity. The space resolution in x and y directions is 1 mm. As seen in Figure 7, these images are characterized as low contrast, intensity inhomogeneity, weak object boundary. Some preprocessing such as filtering and intensity adjustment skull stripping were taken. We employ the Jaccard similarity index (JS) to quantitatively measure the segmentation performance of these methods. This JS index is the ratio between intersectional area of S1 and S2 and their united area; that is,(18)J(S1,S2)=area(S1∩S2)area(S1∪S2).The closer the J is to 1, the more similar S1 is to S2. In our experiments, S1 is the segmented region by the four compared methods, and S2 is the ground truth. For each method, we used its optimal initial contour to obtain their best performance. Other common parameters in these methods were settled as same as possible for fair comparison, except the iteration number and the exclusive parameters of each method. The evolution iteration will stop once the difference of two neighboring contours of the zero level set is less than ten pixels. It should be noted that the larger Δt can speed up the evolution of level sets in these methods but easily makes boundary leakage or zero LS disappear; so we put the segment accuracy as the first place during the selection of these parameters, for example, Δt = 0.05~0.2. Other parameters are searched in ν = 0.001∗2552 ~ 0.003∗2552 with step 0.001∗2552; μ = 0.2/Δt; w = 0.8~0.99 with step 0.05; k = 0.01~10 with step 0.05. To obtain fair experimental results, the programs were conducted 50 times under same condition; then the average of the JS values and CPU time consuming is computed as indices for measuring the efficiency of each method. The compared methods were programmed just based on the original papers and not algorithmic optimization or programming technique was taken into consideration; neither narrow band constriction nor other skills were incorporated into the compared methods.


New region-scalable discriminant and fitting energy functional for driving geometric active contours in medical image segmentation.

Wang X, Niu Y, Tan L, Zhang SX - Comput Math Methods Med (2014)

Testing images in quantitative experiments.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4109108&req=5

fig7: Testing images in quantitative experiments.
Mentions: To quantitatively evaluate the performance of our proposed model, we use it to segment five brain MRI images of ground-truth object (i.e., white matter) and compare it to CV, RSF, and LGIF methods. The 2D MR images are taken from McGill Brain Web [31] with 3% noise level and 40% intensity nonuniformity. The space resolution in x and y directions is 1 mm. As seen in Figure 7, these images are characterized as low contrast, intensity inhomogeneity, weak object boundary. Some preprocessing such as filtering and intensity adjustment skull stripping were taken. We employ the Jaccard similarity index (JS) to quantitatively measure the segmentation performance of these methods. This JS index is the ratio between intersectional area of S1 and S2 and their united area; that is,(18)J(S1,S2)=area(S1∩S2)area(S1∪S2).The closer the J is to 1, the more similar S1 is to S2. In our experiments, S1 is the segmented region by the four compared methods, and S2 is the ground truth. For each method, we used its optimal initial contour to obtain their best performance. Other common parameters in these methods were settled as same as possible for fair comparison, except the iteration number and the exclusive parameters of each method. The evolution iteration will stop once the difference of two neighboring contours of the zero level set is less than ten pixels. It should be noted that the larger Δt can speed up the evolution of level sets in these methods but easily makes boundary leakage or zero LS disappear; so we put the segment accuracy as the first place during the selection of these parameters, for example, Δt = 0.05~0.2. Other parameters are searched in ν = 0.001∗2552 ~ 0.003∗2552 with step 0.001∗2552; μ = 0.2/Δt; w = 0.8~0.99 with step 0.05; k = 0.01~10 with step 0.05. To obtain fair experimental results, the programs were conducted 50 times under same condition; then the average of the JS values and CPU time consuming is computed as indices for measuring the efficiency of each method. The compared methods were programmed just based on the original papers and not algorithmic optimization or programming technique was taken into consideration; neither narrow band constriction nor other skills were incorporated into the compared methods.

Bottom Line: The region-scalable discriminant and fitting energy functional is defined to capture the image intensity characteristics in local and global regions for driving the evolution of active contour.The discriminant term in the model aims at separating background and foreground in scalable regions while the fitting term tends to fit the intensity in these regions.The new model utilizes intensity information in the local and global regions as much as possible; so it not only handles better intensity inhomogeneity, but also allows more robustness to noise and more flexible initialization in comparison to the original global region and regional-scalable based models.

View Article: PubMed Central - PubMed

Affiliation: Key Laboratory of Optoelectronic Technology and Systems of Ministry of Education, College of Optoelectronic Engineering, Chongqing University, Chongqing 400044, China.

ABSTRACT
We propose a novel region-based geometric active contour model that uses region-scalable discriminant and fitting energy functional for handling the intensity inhomogeneity and weak boundary problems in medical image segmentation. The region-scalable discriminant and fitting energy functional is defined to capture the image intensity characteristics in local and global regions for driving the evolution of active contour. The discriminant term in the model aims at separating background and foreground in scalable regions while the fitting term tends to fit the intensity in these regions. This model is then transformed into a variational level set formulation with a level set regularization term for accurate computation. The new model utilizes intensity information in the local and global regions as much as possible; so it not only handles better intensity inhomogeneity, but also allows more robustness to noise and more flexible initialization in comparison to the original global region and regional-scalable based models. Experimental results for synthetic and real medical image segmentation show the advantages of the proposed method in terms of accuracy and robustness.

Show MeSH
Related in: MedlinePlus