Mammography-based artificial intelligence model for predicting axillary lymph node status after neoadjuvant therapy...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Accurate assessment of axillary lymph node (ALN) status after neoadjuvant therapy (NAT) is crucial for breast cancer treatment decisions.
• Current methods lack reliability, leading to potential overtreatment in some patients.

Purpose of the Study:

• To develop and validate a deep learning-based artificial intelligence (AI) model for predicting post-NAT ALN status using digital mammography (DM) images.
• To improve treatment selection and reduce unnecessary surgeries in breast cancer patients.

Main Methods:

• Developed and validated an AI model using DM images and clinical data from 956 invasive breast cancer patients with positive ALN metastasis.
• Compared various image cropping methods and backbone networks (e.g., Swin Transformer V2) for optimal classification architecture.
• Evaluated model performance using ROC curves and AUC on internal and external test sets.

Main Results:

• The AI model utilizing "fixed 5 cm" image clipping and Swin Transformer V2 achieved the best ALN status prediction performance.
• Incorporating pre-training models and clinical features significantly improved prediction accuracy (AUC > 0.8 in all datasets).
• The model demonstrated robust performance across training, internal validation, internal test, and external test sets.

Conclusions:

• An AI model based on baseline DM images can effectively predict ALN status in breast cancer patients after NAT.
• The developed AI model shows promise in providing decision support to clinicians, potentially avoiding excessive surgical interventions.
• This AI tool can aid in selecting more beneficial treatment modalities for individual patients.