Mammographic-Mass-Data-Set

961 mammographic mass lesions (516 benign, 445 malignant) collected at the Institute of Radiology of the University Erlangen-Nuremberg between 2003 and 2006. The dataset includes BI-RADS assessment, patient age, and three BI-RADS attributes (shape, margin, density) to predict lesion severity. It can be used to evaluate computer-aided diagnosis (CAD) systems against radiologist assessments.

Use Cases

• Train binary classifiers to predict benign vs. malignant lesions based on BI-RADS attributes and patient age.
• Benchmark CAD system performance by comparing model sensitivity and specificity to radiologist BI-RADS assessments.
• Analyze the relationship between specific mammographic features (shape, margin, density) and malignancy risk.
• Develop decision-support tools to help reduce unnecessary breast biopsies by improving positive predictive value.

Strengths

• 961 real-world clinical instances with a clear class distribution (516 benign, 445 malignant).
• Includes ground truth severity labels and BI-RADS assessments from a double-review process by physicians.
• Data collection is documented with a specific time range (2003-2006) and institution.

Limitations

• Contains missing values across several attributes (e.g., 76 missing for Density, 48 for Margin).
• Row count and file format details are unknown from the provided metadata.
• The dataset's age (2003-2006) may limit its applicability to current mammography practices.

Provenance

Source

Institute of Radiology of the University Erlangen-Nuremberg

Collection Method

Collected from full field digital mammograms, with BI-RADS assessments assigned in a double-review process.

Time Range

2003 to 2006