BUS-UCLM: Breast ultrasound lesion segmentation dataset

Area of Science:

• Medical Imaging
• Computer Vision
• Public Health

Background:

• Breast cancer diagnosis relies on accurate lesion identification from ultrasound images.
• Segmentation and classification of breast lesions are crucial for determining malignancy.
• Existing datasets may lack detailed annotations for robust machine learning model development.

Purpose of the Study:

• To present a comprehensive dataset of breast ultrasound scans for lesion segmentation and classification.
• To provide expert-annotated ground truth masks for training and evaluating computer vision models.
• To facilitate research in automated detection and characterization of breast tumors.

Main Methods:

• Collection of 38 breast ultrasound scans (683 images) using a Siemens ACUSON S2000TM Ultrasound System.
• Categorization of images into normal (419), benign (174), and malignant (90) findings.
• Generation of RGB segmentation masks as ground truth, with specific colors denoting tissue types and lesion classifications.

Main Results:

• A dataset with 683 images, including detailed segmentation masks for normal tissue, benign lesions, and malignant lesions.
• The dataset is suitable for training and validating machine learning algorithms for breast lesion analysis.
• Enables quantitative assessment of lesion area and contour through segmentation.

Conclusions:

• This dataset is a valuable resource for advancing research in breast lesion segmentation and classification.
• It supports the development of AI-driven tools for improved breast cancer diagnosis.
• Facilitates public health initiatives through enhanced medical imaging analysis.