Predicting axillary lymph node metastasis in breast cancer using a multimodal radiomics and deep learning model

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate prediction of axillary lymph node metastasis (ALNM) is crucial for breast cancer (BC) management.
• Integrating imaging data with machine learning can enhance diagnostic capabilities.

Purpose of the Study:

• To evaluate combined radiomics and deep learning models for ALNM prediction in BC.
• To compare the performance of different machine learning algorithms in these models.
• To guide individualized treatment plans and preoperative interventions.

Main Methods:

• Retrospective analysis of 270 BC patients' mammography (MG) and MRI data.
• Extraction and fusion of radiomics and deep learning (3D-Resnet18) features.
• Development and evaluation of radiomics, deep learning, and combined models using machine learning algorithms and AUC analysis.

Main Results:

• The combined radiomics-deep learning model using Multilayer Perceptron (MLP) achieved the highest Area Under the Curve (AUC) of 0.846.
• The combined-MLP model outperformed individual radiomics (AUC=0.756) and deep learning (AUC=0.712) models.
• MLP demonstrated strong classification performance in predicting ALNM.

Conclusions:

• Multimodal radiomics and deep learning models offer significant value for preoperative ALNM prediction in BC.
• These models can aid in developing scientifically individualized treatment plans.
• The findings support the use of advanced AI models in clinical decision-making for breast cancer.