Molecular-subtype guided automatic invasive breast cancer grading using dynamic contrast-enhanced MRI

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Histological grade and molecular subtype are key prognostic indicators for invasive breast cancer (IBC).
• Personalized medicine relies on accurate prognostic markers for invasive breast cancer (IBC).
• Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) offers insights into tumor biology.

Purpose of the Study:

• To evaluate a two-stage deep learning framework for invasive breast cancer (IBC) grading.
• To incorporate molecular subtype (MS) information into deep learning for enhanced IBC grading.
• To assess the efficacy of DCE-MRI in preoperative breast cancer assessment.

Main Methods:

• Developed a two-stage deep learning model (IOS²-DA) integrating molecular subtype (MS) data.
• Stage I utilized a novel neural network (IOS²-DA) focusing on imaging manifestations of IBC grades.
• Stage II incorporated an MS attention branch to refine predictions using Kullback-Leibler divergence and ensemble learning.

Main Results:

• The MS-guided IOS²-DA significantly outperformed the single model, achieving high accuracy (0.927), precision (0.942), AUC (0.927), and F1-score (0.930).
• Gradient-weighted class activation maps confirmed that extracted features align with tumor areas.
• The DeLong test indicated statistically significant improvements (P < 0.05).

Conclusions:

• The IOS²-DA framework shows potential for non-invasive invasive breast cancer (IBC) tumor grade prediction.
• Integrating molecular subtype (MS) information enhances diagnostic effectiveness for IBC grading.
• DCE-MRI is a feasible imaging modality for comprehensive preoperative assessment of breast cancer biology and prognosis.