Automatic hip cartilage segmentation from 3D MR images using arc-weighted graph searching

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Radiology

Background:

• Accurate segmentation of hip joint cartilage is essential for quantitative analysis of osteoarthritis.
• Current segmentation methods may struggle with weak or incomplete boundaries in MR images.

Purpose of the Study:

• To develop and validate a fully automatic scheme for segmenting individual femoral and acetabular cartilage plates in 3D MR images.
• To improve quantitative investigations of hip joint pathoanatomical conditions.

Main Methods:

• Utilized an improved optimal multi-object multi-surface graph search framework with arc-weighted graph representation.
• Incorporated prior morphological knowledge for robust segmentation.
• Validated against manual segmentations from 52 asymptomatic volunteers' 3D TrueFISP MR images.

Main Results:

• Achieved mean Dice's similarity coefficients of 0.81 (±0.03) for combined, 0.79 (±0.03) for femoral, and 0.72 (±0.05) for acetabular cartilage volumes.
• Reported mean absolute volume difference errors of 8.44% (±6.36), 9.44% (±7.19), and 9.05% (±8.02) respectively.
• Demonstrated mean absolute cartilage thickness differences of 0.13 mm (±0.12) for femoral and 0.11 mm (±0.11) for acetabular plates.

Conclusions:

• The developed automated scheme provides accurate and reliable segmentation of hip joint cartilage.
• This method facilitates quantitative analysis for conditions like osteoarthritis.
• The approach effectively handles challenging MR image data with weak boundaries.