A deep learning framework to stratify Nottingham histologic grade 2 breast tumors based on dynamic contrast-enhanced...

Area of Science:

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

Background:

• The Nottingham Histologic Grade (NHG) is crucial for breast cancer prognosis and treatment.
• NHG2 tumors exhibit significant biological heterogeneity, complicating treatment decisions and leading to potential overtreatment or undertreatment.

Purpose of the Study:

• To develop and validate a deep learning model (DeepRadGrade) for stratifying NHG2 breast tumors.
• To assess the clinical utility of this model in predicting recurrence-free survival (RFS).

Main Methods:

• A convolutional neural network (CNN) was trained on dynamic contrast-enhanced (DCE) MRI data to distinguish NHG1 from NHG3 tumors.
• The trained model classified 456 NHG2 tumors into NHG1-like (DRG2-) and NHG3-like (DRG2+) subgroups.
• Recurrence-free survival was analyzed using Kaplan-Meier and Cox models, adjusting for standard prognostic factors.

Main Results:

• DeepRadGrade (DRG) demonstrated strong performance in differentiating tumor grades across training, testing, and external validation datasets (AUCs ranging from 0.82 to 0.84).
• Among NHG2 tumors, 315 were classified as DRG2- and 131 as DRG2+.
• Patients with DRG2+ tumors showed significantly worse RFS (adjusted hazard ratio = 2.39, p=0.0059), indicating higher recurrence risk.
• The inclusion of DRG classification improved the predictive accuracy of the Cox model (C-index increased from 0.68 to 0.73).

Conclusions:

• Deep learning applied to routine DCE MRI can effectively stratify NHG2 breast tumors into clinically meaningful subgroups based on recurrence risk.
• This AI-driven approach provides a cost-effective method for individualized risk stratification in intermediate-grade breast cancer.
• The findings suggest that DRG classification can help optimize treatment strategies, minimizing both under- and overtreatment.