A machine learning approach for differentiating malignant from benign enhancing foci on breast MRI

Area of Science:

• Radiology
• Medical Imaging
• Machine Learning

Background:

• Differentiating malignant from benign enhancing foci on breast MRI is crucial for accurate diagnosis.
• Radiomic signatures offer a potential non-invasive method for lesion characterization.

Purpose of the Study:

• To evaluate the feasibility of using radiomic features and machine learning to distinguish malignant from benign enhancing foci on breast MRI.
• To assess the diagnostic performance of a radiomic approach in breast lesion characterization.

Main Methods:

• A retrospective study included 45 enhancing foci from 45 patients, with additional reference cases.
• Over 200 radiomic features were extracted from MRI scans and analyzed using an evolutionary machine learning method.
• A k-nearest neighbour classifier was developed and evaluated for sensitivity, specificity, and accuracy.

Main Results:

• The k-nearest neighbour classifier, utilizing 35 selected features, demonstrated high performance.
• The best performing classifier achieved 100% sensitivity, 90% specificity, and 94% accuracy in differentiating lesions.
• These results were based on a combined dataset of 68 cases.

Conclusions:

• This preliminary study confirms the feasibility of employing a radiomic approach for characterizing enhancing foci on breast MRI.
• The findings suggest that radiomics can be a valuable tool in improving the diagnostic accuracy of breast MRI.