Deep Learning Models Based on Pretreatment MRI and Clinicopathological Data to Predict Responses to Neoadjuvant...

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Triple-negative breast cancer (TNBC) presents a significant challenge in oncologic treatment.
• Predicting pathologic complete response (pCR) to neoadjuvant systemic therapy (NAST) is crucial for treatment stratification.
• Multiparametric breast MRI and clinicopathological data offer rich information for predictive modeling.

Purpose of the Study:

• To develop and validate deep learning models for predicting pCR to NAST in TNBC patients.
• To assess the utility of pretreatment multiparametric breast MRI and clinicopathological data in predictive model development.
• To investigate the impact of various model parameters on prediction performance.

Main Methods:

• A prospective study of 282 TNBC patients (Stage I-III) using baseline multiparametric MRI (DCE, DWI) and clinicopathological data.
• Development and internal testing of 48 distinct 3D deep learning models.
• External validation using data from the I-SPY 2 trial.
• Systematic investigation of 3D model frameworks, tumor volume preprocessing, ROI selection, and data inputs.

Main Results:

• The best internal model, integrating DCE, DWI, and clinicopathological data, achieved an AUC of 0.76.
• External validation demonstrated strong performance, with models using only DCE or DWI achieving an AUC of 0.72.
• Model performance was influenced by choices in 3D frameworks, tumor volume processing, and data inputs.

Conclusions:

• 3D deep learning models utilizing pretreatment data can effectively predict pCR to NAST in TNBC.
• These models hold promise for enhancing personalized treatment strategies in TNBC management.
• Further research can refine these models for improved clinical applicability.