Predicting Breast Cancer Subtypes Using Magnetic Resonance Imaging Based Radiomics With Automatic Segmentation

Area of Science:

• Medical Imaging
• Oncology
• Artificial Intelligence

Background:

• Accurate breast cancer molecular subtyping is crucial for guiding treatment decisions.
• Current methods for subtyping can be invasive or time-consuming.
• There is a need for noninvasive, efficient methods to predict molecular subtypes.

Purpose of the Study:

• To evaluate the feasibility of radiomics, utilizing automatic segmentation, for predicting breast cancer molecular subtypes.
• To assess the performance of radiomics models in classifying different molecular subtypes.

Main Methods:

• A retrospective study of 516 breast cancer patients was conducted.
• An automatic 3D UNet-based Convolutional Neural Network was used for region of interest segmentation.
• 1316 radiomics features were extracted, and 18 radiomics methods were employed for model selection and performance assessment using AUC, accuracy, sensitivity, and specificity.

Main Results:

• The automatic segmentation achieved an average Dice similarity coefficient of 0.89.
• Radiomics models successfully predicted 4 molecular subtypes with an average AUC of 0.8623.
• High performance was observed for specific subtype predictions, including triple-negative breast cancer (AUC = 0.9335).

Conclusions:

• Radiomics based on automatic MRI segmentation offers a noninvasive approach for predicting breast cancer molecular subtypes.
• This method demonstrates potential for application in large patient cohorts for improved breast cancer management.