Active volume models with probabilistic object boundary prediction module.

Area of Science:

• Medical image analysis
• Computer vision
• Computational modeling

Background:

• Traditional active contours (Snakes, level-sets) focus on curves/surfaces, limiting their ability to model volumetric objects.
• Accurate segmentation and tracking of objects in noisy 2D/3D medical images remain challenging.

Purpose of the Study:

• To propose a novel Active Volume Model (AVM) for free-form object segmentation and tracking.
• To enhance segmentation by incorporating both boundary and interior object information.
• To improve upon existing methods by integrating region-based constraints with the efficiency of active contour models.

Main Methods:

• The Active Volume Model (AVM) deforms to minimize energy, incorporating both boundary and interior object properties.
• The model alternates between deformation based on object prediction and appearance statistics.
• A probabilistic object prediction module uses the Bayesian Decision Rule for foreground/background separation.
• Model optimization extends the Snakes model, integrating region information into external forces.

Main Results:

• The AVM demonstrates efficient segmentation with adaptive region-based constraints.
• Experimental results on noisy 2D/3D medical images show competitive performance.
• Comparisons with GVF Snakes and level set methods validate the AVM's effectiveness.

Conclusions:

• The Active Volume Model (AVM) offers an efficient and adaptive approach for object segmentation and tracking in medical imaging.
• AVM effectively integrates region information, overcoming limitations of curve/surface-based active contours.
• The model shows promise for applications involving noisy 2D/3D medical image data.