Automatic kidney segmentation in CT images based on multi-atlas image registration

Area of Science:

• Medical Imaging
• Urology
• Computer-Aided Diagnosis

Background:

• Accurate kidney segmentation is crucial for urological computer-aided diagnosis and treatment planning.
• Existing methods may lack the precision required for clinical applications.

Purpose of the Study:

• To present an automatic, multi-atlas image registration-based method for precise kidney segmentation.
• To evaluate the accuracy and feasibility of the proposed method on CT angiographic (CTA) and CT urography (CTU) images.

Main Methods:

• A two-step coarse-to-fine framework utilizing multi-atlas image registration.
• Step 1: Registration of down-sampled patient images with low-resolution atlases for initial kidney localization.
• Step 2: Cropping and alignment of kidneys with high-resolution atlases for refined segmentation.

Main Results:

• Achieved high accuracy in kidney segmentation on 14 CTA images, with an average Dice similarity coefficient of 0.952.
• Demonstrated a low average surface-to-surface distance of 0.913mm between segmented and reference kidneys.
• Showcased the method's feasibility for kidney segmentation in both CTA and CTU images from 12 patients.

Conclusions:

• The proposed automatic multi-atlas registration method provides accurate and reliable kidney segmentation.
• This technique holds significant potential for enhancing computer-aided diagnosis and treatment in urology.
• The method is effective across different CT imaging modalities, including CTA and CTU.