Accurate bone segmentation in 2D radiographs using fully automatic shape model matching based on regression-voting

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Shape model matching is crucial for analyzing anatomical structures in medical images.
• Previous methods often require manual intervention or lack robustness.

Purpose of the Study:

• To apply Random Forests (RFs) regression-voting within a fully automatic shape model matching (FASMM) system.
• To evaluate the system's performance on diverse radiograph segmentation tasks.
• To elucidate the key properties contributing to the approach's effectiveness.

Main Methods:

• Utilized RF regression-voting for feature point localization in shape model matching.
• Applied the FASMM system to segment the proximal femur, knee joint, and hand joints in radiographs.
• Investigated the impact of vote integration, ensemble voting, and coarse-to-fine strategies.

Main Results:

• The FASMM system achieved state-of-the-art performance across all three segmentation problems.
• Each investigated property (vote integration, ensemble voting, coarse-to-fine strategy) individually improved performance.
• Combining all three properties yielded the best results, demonstrating superior accuracy and efficiency.

Conclusions:

• RF regression-voting is a highly effective technique for robust and accurate automatic shape model matching in radiographs.
• The FASMM system demonstrates excellent generalizability across different anatomical regions.
• This approach offers an accurate and time-efficient solution for segmenting bony structures in radiographic images.