Chestwall segmentation in 3D breast ultrasound using a deformable volume model

Area of Science:

• Medical imaging
• Biomedical engineering
• Image segmentation

Background:

• Segmentation of breast parenchyma in 3D whole breast ultrasound is challenging due to noise and ill-defined boundaries.
• Deformable surface models struggle with image artifacts like thoracic shadowing.
• Accurate segmentation is crucial for computer-aided detection (CAD) systems to reduce false positives.

Purpose of the Study:

• To propose and evaluate a deformable volume segmentation method for detecting breast parenchyma in 3D whole breast ultrasound.
• To develop a robust deformable ultrasound volume model incorporating anatomical structures and artifacts.
• To enhance chestwall detection accuracy using a novel rib shadow enhancement filter.

Main Methods:

• A parameterized deformable ultrasound volume model was created, including breast, rib, intercostal space, and thoracic shadowing.
• The model was optimized to match ultrasound scans using prior knowledge of grey value statistics and shape.
• A Hessian sheet detector-based rib shadow enhancement filter was developed and integrated.

Main Results:

• The deformable volume segmentation method demonstrated potential for breast parenchyma extraction.
• An ROC study on 88 multi-center scans showed significant chestwall detection accuracy.
• The integrated rib shadow enhancement filter notably improved chestwall detection performance.

Conclusions:

• Deformable volume segmentation is a viable alternative to surface models for challenging ultrasound images.
• The proposed method effectively segments breast parenchyma and improves chestwall detection.
• This approach could enhance the performance of computer-aided lesion detection by reducing false positives.