Automatic Segmentation and Molecular Subtype Classification of Breast Cancer Using an MRI-based Deep Learning...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Oncology
• Breast Cancer Research

Background:

• Accurate molecular subtyping of breast cancer is crucial for treatment selection and prognosis.
• Current methods for subtyping can be invasive and time-consuming.
• Deep learning offers potential for automated analysis of medical imaging data.

Purpose of the Study:

• To develop a deep learning framework utilizing contrast-enhanced MRI for automated breast cancer lesion segmentation.
• To achieve automatic classification of breast cancer molecular subtypes using the developed framework.
• To evaluate the framework's performance across multiple datasets for robustness.

Main Methods:

• A retrospective multicenter study included 687 patients with invasive breast cancer.
• A 3D ResU-Net model was employed for automatic breast lesion segmentation.
• An ensemble model (Ensemble ResNet) combined 2D and 3D features for molecular subtype classification.

Main Results:

• The segmentation model achieved high Dice scores (0.82-0.86) across internal and external datasets.
• The Ensemble ResNet model demonstrated strong performance in classifying Luminal A, Luminal B, HER2-enriched, and Triple-Negative Breast Cancer (TNBC) subtypes (AUC range 0.68-0.84).
• The framework showed robust and accurate classification across all tested datasets.

Conclusions:

• The proposed deep learning framework based on MRI provides a highly accurate and robust method for fully automatic breast cancer molecular subtype classification.
• This AI-driven approach has the potential to significantly improve breast cancer diagnosis and management.
• The framework's performance highlights the utility of advanced imaging and AI in precision oncology.