3D Fast Automatic Segmentation of Kidney Based on Modified AAM and Random Forest

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Renal Anatomy

Background:

• Accurate kidney segmentation is crucial for diagnosing and treating renal diseases.
• Existing segmentation methods often lack automation and precision, particularly in 3D CT data.
• The complexity of renal structures necessitates advanced segmentation techniques.

Purpose of the Study:

• To develop a fully automatic and fast method for segmenting kidneys into four distinct components: renal cortex, renal column, renal medulla, and renal pelvis.
• To evaluate the accuracy and efficiency of the proposed segmentation method on clinical 3D CT abdominal images.

Main Methods:

• A two-part approach combining 3D Generalized Hough Transform (GHT) and 3D Active Appearance Models (AAM) for renal cortex localization.
• A modified Random Forests (RF) method for segmenting the kidney into four components, leveraging the localization results.
• Implementation of multithreading technology to accelerate the segmentation process.

Main Results:

• The method achieved high true-positive volume fractions: 93.15% for renal cortex, 83.09% for renal column, 81.92% for renal medulla, and 80.28% for renal pelvis.
• Low false-positive volume fractions were reported across all segmented components, indicating high specificity.
• The average computational time for segmenting the kidney into four components was approximately 20 seconds.

Conclusions:

• The proposed fully automatic method offers an efficient and accurate solution for segmenting kidneys into multiple components from 3D CT images.
• This technique has the potential to significantly aid in clinical diagnosis and treatment planning for renal pathologies.
• The combination of GHT, AAM, and RF, along with multithreading, demonstrates a robust approach to medical image segmentation.