Develop and Validate a Nomogram Combining Contrast-Enhanced Spectral Mammography Deep Learning with...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Accurate prediction of neoadjuvant chemotherapy (NAC) response is crucial for managing ER-positive/HER2-negative breast cancer.
• Current prediction methods may not fully leverage advanced imaging and clinical data.

Purpose of the Study:

• To develop and validate a nomogram integrating contrast-enhanced spectral mammography (CESM) deep learning features with clinical-pathological data.
• To predict NAC response (low vs. high Miller Payne grades) in ER-positive/HER2-negative breast cancer patients.

Main Methods:

• A retrospective study of 265 patients with ER-positive/HER2-negative breast cancer.
• Deep learning models (ResNet34) trained on CESM images, combined with clinical-pathological features (age, PR, ER, Ki67, neutrophil-to-lymphocyte ratio).
• Nomogram development using logistic regression and stepwise selection; evaluation via ROC curves, precision-recall curves, and decision curve analysis (DCA).

Main Results:

• The nomogram demonstrated superior predictive performance compared to deep learning-based and clinical models.
• Key metrics for the nomogram included high area under the ROC curve (0.95), accuracy (0.94), specificity (0.98), and positive predictive value (0.89).
• Decision curve analysis indicated significant clinical utility for the nomogram in treatment decision-making.

Conclusions:

• The developed nomogram effectively predicts NAC response in ER-positive/HER2-negative breast cancer.
• Integrating CESM deep learning with clinical-pathological features enhances predictive accuracy.
• The nomogram shows promise as a valuable tool for personalized breast cancer treatment strategies.