Multimodal Breast Ultrasound Images for Malignancy Risk Stratification

3,703 ultrasound images from 2,685 patients were used to develop machine learning models for breast cancer diagnosis. The dataset includes 2,069 benign and 616 malignant cases collected between July 2019 and March 2024. Shengxin Pei authored this research, which compares models using BI-RADS terminology, ultrasound imaging, and radiomics features.

Use Cases

• Training machine learning models for breast mass malignancy classification based on combined multimodal features.
• Comparing diagnostic performance of Logistic Regression, SVM, and Random Forest models across BI-RADS categories.
• Validating radiomics feature extraction pipelines for ultrasound imaging data.
• Studying model performance in specific BI-RADS subcategories (2, 3, 4a, 4b, 4c, 5) to identify clinical utility gaps.

Strengths

• Dataset contains 3,703 ultrasound images from 2,685 patients, providing a substantial sample size.
• Includes clear pathological diagnosis labels for 2,069 benign and 616 malignant cases.
• Data is split into training (2,979 images) and validation (724 images) sets, supporting model development workflows.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data is from a single medical center, which may limit generalizability.
• Row count for the underlying feature data is unknown, which may limit suitability assessment.

Provenance

Source

figshare

Collection Method

Retrospective cohort study analyzing ultrasound images and patient data.

Time Range

July 2019 to March 2024

Freshness

Last updated 2026-06-03 04:47:28; freshness should be verified.