Segmentation of abdomen MR images using kernel graph cuts with shape priors

Area of Science:

• Medical Image Analysis
• Computer Vision
• Biomedical Engineering

Background:

• Abdominal organ segmentation in Magnetic Resonance (MR) imaging presents significant challenges.
• Issues include intensity inhomogeneity, weak boundaries, noise, and the presence of similar adjacent tissues.

Purpose of the Study:

• To propose a novel and accurate method for abdominal organ segmentation in MR images.
• To address limitations of existing segmentation techniques, particularly boundary leakage and errors caused by similar tissues.

Main Methods:

• A hybrid approach combining Kernel Graph Cuts (KGC) with shape priors.
• Initial contour generation using region growing and morphology operations.
• Shape priors derived using Kernel Principal Component Analysis (KPCA) on registered shape templates.
• Integration of shape priors into the KGC energy function for robust segmentation.

Main Results:

• The proposed method achieved satisfying segmentation of abdominal organs (liver, kidneys) without boundary leakage or errors from similar tissues.
• Quantitative comparison using Probabilistic Rand Index (PRI) and Variation of Information (VoI) demonstrated superior performance.
• Achieved highest PRI (up to 0.9983) and lowest VoI (down to 0.3205) compared to other methods.

Conclusions:

• The novel KGC with shape priors method effectively overcomes boundary leakage and segmentation errors in abdominal MR images.
• Integration of KPCA-based shape priors enhances the robustness and accuracy of the graph cuts algorithm.
• The method provides reliable segmentation even when dealing with challenging image artifacts and tissue similarities.