Watershed segmentation for breast tumor in 2-D sonography

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Accurate breast tumor contouring in ultrasound (US) imaging is crucial for diagnosis, especially for physicians lacking extensive experience.
• Current manual contouring can be time-consuming and subjective, highlighting the need for automated solutions.

Purpose of the Study:

• To develop and evaluate an automated method for precise breast tumor contouring using medical ultrasound images.
• To integrate neural network (NN) classification with morphological watershed segmentation for enhanced accuracy.

Main Methods:

• Utilized textural analysis with autocovariance coefficients as input features for a neural network (NN) classifier.
• Employed a self-organizing map (SOM) for texture classification and adaptive preprocessing selection.
• Applied watershed transformation for automatic tumor contour segmentation.

Main Results:

• The proposed automated method demonstrated consistent identification of tumor contours and regions-of-interest (ROIs) comparable to manual contouring by experienced physicians.
• Computer simulations confirmed the robustness and speed of the automatic contouring technique.
• The method achieved high stability in segmenting breast tumors from ultrasound images.

Conclusions:

• The developed automated contouring method offers a reliable and efficient tool for breast tumor analysis in ultrasound imaging.
• This technique can significantly reduce the time required for segmentation in computer-aided diagnosis (CAD) systems.
• The automated approach supports physicians by providing fast and stable contouring, complementing manual assessments.