3D kidney segmentation from CT images using a level set approach guided by a novel stochastic speed function

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Renal Imaging

Background:

• Accurate kidney segmentation is crucial for noninvasive computer-aided diagnosis (CAD) systems.
• Early detection of acute renal rejection relies on precise segmentation of kidney structures.
• Computed Tomography (CT) imaging is a common modality for visualizing abdominal organs.

Purpose of the Study:

• To introduce a novel 3-D segmentation approach for kidneys from CT images.
• To enhance the accuracy and robustness of kidney segmentation for CAD applications.
• To facilitate early detection of acute renal rejection through improved segmentation.

Main Methods:

• A geometric deformable model guided by a stochastic speed relationship was employed.
• A two-level joint Markov-Gibbs random field (MGRF) model integrated shape priors and appearance features.
• Voxel-wise image intensities and their spatial interactions were utilized for segmentation.

Main Results:

• The proposed 3-D segmentation approach demonstrated robustness and accuracy.
• Performance was evaluated on 21 CT datasets with manual expert segmentation.
• Receiver Operating Characteristic (ROC) and Dice Similarity Coefficient (DSC) confirmed high performance.

Conclusions:

• The developed segmentation method is accurate and robust for kidney segmentation from CT images.
• This approach can significantly contribute to the development of noninvasive CAD systems for renal disease.
• The method shows promise for improving the early detection of acute renal rejection.