3D medical image segmentation by multiple-surface active volume models.

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Anatomy

Background:

• Accurate 3D object extraction from medical images is crucial for diagnosis and treatment planning.
• Existing Active Volume Models (AVM) can be limited by local minima and leakage, affecting segmentation accuracy.
• Incorporating spatial relationships between multiple objects can enhance segmentation robustness.

Purpose of the Study:

• To introduce Multiple-Surface Active Volume Models (MSAVM) for improved 3D object extraction from volumetric medical images.
• To enhance the accuracy and robustness of 3D medical image segmentation compared to original Active Volume Models.
• To develop a method that effectively integrates image-based region information with spatial constraints among multiple surfaces.

Main Methods:

• Developed MSAVM, an extension of Active Volume Models (AVM), incorporating spatial constraints among multiple objects.
• Introduced two novel surface-distance based functions to adaptively adjust contributions from image data and spatial constraints.
• Utilized the implicit representation of AVM and signed distance transform maps for efficient spatial information calculation.

Main Results:

• MSAVM demonstrated increased robustness and accuracy in 3D object extraction compared to original AVM.
• The proposed adaptive functions effectively addressed issues like local minima and segmentation leakage.
• Experiments on volumetric medical images validated the superior performance of MSAVM.

Conclusions:

• MSAVM offers a more accurate and robust solution for 3D object segmentation in medical imaging.
• The method efficiently integrates spatial information with image data, overcoming limitations of previous models.
• MSAVM provides a computationally efficient yet highly accurate approach for volumetric medical image analysis.