Accurate fully automatic femur segmentation in pelvic radiographs using regression voting

Area of Science:

• Medical Imaging
• Radiography Analysis
• Biomedical Engineering

Background:

• Accurate bone contour extraction from radiographs is crucial for medical diagnosis, surgical planning, and treatment assessment.
• Current methods may lack the precision or automation required for efficient clinical workflows.

Purpose of the Study:

• To develop and validate a fully automated method for precise segmentation of the proximal femur in anteroposterior pelvic radiographs.
• To establish a new benchmark for accuracy in automated radiographic bone segmentation.

Main Methods:

• A novel approach combining global search with a detector to identify candidate positions.
• Refinement of candidate positions using a statistical shape model and local detectors.
• Implementation of Random Forest regression for both global and local model optimization.

Main Results:

• The automated system achieved a mean point-to-curve error below 1 mm for 98% of the 519 evaluated radiographs.
• Demonstrated robust and accurate segmentation of the proximal femur.
• Outperformed existing automated methods in accuracy.

Conclusions:

• The developed fully automated method offers high accuracy and robustness for proximal femur segmentation in pelvic radiographs.
• This technique has the potential to significantly enhance diagnostic capabilities and pre-operative planning in orthopedics.
• Represents a significant advancement in automated medical image analysis for skeletal structures.