Deep Learning Model Based on Dual-Modal Ultrasound and Molecular Data for Predicting Response to Neoadjuvant...

Area of Science:

• Medical Imaging
• Oncology
• Artificial Intelligence

Background:

• Predicting neoadjuvant chemotherapy (NAC) response in breast cancer is crucial for treatment selection.
• Current methods often lack precision, necessitating improved non-invasive biomarkers.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (CNN) model for predicting NAC response in breast cancer patients.
• To compare the efficacy of radiomics and CNN models using B-mode ultrasound (BUS) and shear wave elastography (SWE).

Main Methods:

• Prospective study of 255 breast cancer patients receiving NAC.
• Development of radiomics and CNN (ResNet) models using pre-treatment BUS and SWE data.
• Integration of dual-modal ultrasound data with clinicopathologic characteristics for a final predictive model.

Main Results:

• Pretreatment SWE demonstrated superior predictive performance over BUS for both radiomics and CNN models.
• CNN models outperformed radiomics models, with higher AUCs for both BUS (0.72 vs. 0.69) and SWE (0.80 vs. 0.77).
• The CNN model integrating dual-modal ultrasound and molecular data achieved high accuracy (83.60%), sensitivity (87.76%), and specificity (77.45%).

Conclusions:

• A pretreatment CNN model utilizing dual-modal ultrasound and molecular data shows excellent performance in predicting NAC response.
• This non-invasive model holds potential as an objective biomarker to guide individualized breast cancer treatment strategies.