Radiomics-based mammographic abnormality identification via radiologist annotations.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiomics

Background:

• Screening mammography is crucial for early breast cancer detection.
• Distinguishing subtle abnormalities from normal tissue remains a challenge.
• Radiomics offers a quantitative approach to analyze medical images.

Purpose of the Study:

• To develop and evaluate a radiomics-based pipeline for identifying suspicious findings on 2D screening mammograms.
• To train machine learning models to differentiate between radiologist-annotated abnormalities and normal breast tissue.

Main Methods:

• A retrospective study analyzed 1604 screening mammograms from 1294 participants.
• Radiomics features were extracted from regions of interest (ROIs) with abnormalities and normal tissue.
• Multiple machine learning classifiers were trained and evaluated using the area under the receiver operating characteristic curve (AUC).

Main Results:

• The radiomics pipeline achieved AUC values ranging from 0.69 to 0.73.
• No significant differences were observed between the performance of different machine learning models.
• The highest nominal performance (AUC: 0.73) was achieved using ANOVA F-score feature selection and Discriminant Analysis (DA).

Conclusions:

• The developed radiomics pipeline demonstrates potential in differentiating abnormalities from normal tissue on screening mammograms.
• Radiomics shows promise for enhancing breast cancer detection and integrating advanced machine learning into screening workflows.