Fast segmentation of kidney components using random forests and ferns

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Accurate segmentation of kidney components is crucial for medical diagnosis and treatment planning.
• Existing automatic segmentation methods often lack speed or precision.

Purpose of the Study:

• To develop a fast and accurate automatic method for segmenting four key kidney components: renal cortex, renal medulla, renal column, and renal pelvis.
• To evaluate the feasibility and efficacy of a novel segmentation approach combining random forests and random ferns.

Main Methods:

• A coarse-to-fine strategy was employed, integrating random forests and random ferns for initial segmentation.
• A weighted forests-ferns approach with novel potential energy features was used for fine segmentation of kidney components.
• The method was validated on 37 contrast-enhanced CT images, assessing accuracy using Dice similarity coefficient (DSC), true positive volume fraction (TPVF), and false positive volume fraction (FPVF).

Main Results:

• The proposed method achieved high accuracy, with mean DSC values of 89.85% (cortex), 80.60% (column), 86.63% (medulla), and 77.75% (pelvis).
• Segmentation of all four kidney components was completed in approximately 3 seconds.
• The method demonstrated high efficiency and accuracy in segmenting kidney structures.

Conclusions:

• The developed automatic kidney component segmentation method is feasible and effective.
• The combination of random forests and ferns, along with novel potential energy features, enhances segmentation performance.
• The method's high accuracy and speed make it suitable for clinical applications.