Medical breast ultrasound image segmentation by machine learning

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Oncology imaging

Background:

• Breast cancer is a leading cause of cancer diagnosis in women.
• Accurate segmentation of breast ultrasound images is crucial for diagnosis and treatment monitoring.
• Manual segmentation is subjective, time-consuming, and requires expertise.

Purpose of the Study:

• To develop an automated method for segmenting breast ultrasound images.
• To segment four major tissue types: skin, fibroglandular tissue, mass, and fatty tissue.
• To evaluate the performance of a convolutional neural network (CNN) for this task.

Main Methods:

• Utilized a convolutional neural network (CNN) for image segmentation.
• Applied the method to three-dimensional (3D) breast ultrasound images.
• Segmented images into skin, fibroglandular tissue, mass, and fatty tissue.

Main Results:

• Achieved Accuracy, Precision, Recall, and F1-measure over 80% for segmentation.
• Demonstrated a Jaccard similarity index (JSI) of 85.1%, outperforming previous watershed algorithm methods (74.54%).
• The CNN method effectively distinguishes between major functional tissues in breast ultrasound images.

Conclusions:

• The proposed CNN method shows significant potential for assisting in clinical breast cancer diagnosis.
• Automated segmentation can improve the efficiency and objectivity of breast ultrasound analysis.
• This technique may enhance imaging across various medical ultrasound applications.