Applying Data Mining Techniques to Improve Breast Cancer Diagnosis

Area of Science:

• Medical Imaging
• Computational Biology
• Oncology

Background:

• Advanced computer-aided diagnosis systems are crucial for reducing false positives in breast cancer detection.
• Data mining offers a powerful approach to support oncologists in breast cancer classification and diagnosis.

Purpose of the Study:

• To compare two breast cancer datasets for improved diagnostic accuracy.
• To identify optimal data mining methods for predicting benign/malignant lesions, classifying breast density, and distinguishing between masses and microcalcifications.

Main Methods:

• Texture feature extraction using two matrices implemented in Matlab.
• Classification of extracted features using data mining algorithms within the WEKA software.

Main Results:

• Accuracies ranged from 89.3% to 64.7% for benign/malignant lesion prediction.
• Breast density classification achieved accuracies between 75.8% and 78.3%.
• Finding identification (mass/microcalcification) accuracies ranged from 71.0% to 83.1%.

Conclusions:

• Naive Bayes demonstrated superior performance in identifying mass textures.
• Random Forests proved to be a highly effective classifier for most tested breast cancer diagnostic tasks.