Deep Learning Prediction of Axillary Lymph Node Metastasis in Breast Cancer Patients Using Clinical...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate detection of axillary lymph node (ALN) metastases is critical for breast cancer staging and treatment.
• Computed tomography (CT) imaging is a key tool for preoperative assessment.

Purpose of the Study:

• To develop and evaluate a deep learning model for enhanced prediction of ALN metastasis in breast cancer patients.
• To investigate the impact of clinical implication-applied preprocessing and cropping methods on model performance.

Main Methods:

• Utilized 1128 axial CT images from 523 breast cancer patients.
• Employed a CT image preprocessing protocol with clinical implications and two cropping methods (fixed size crop and adjustable square crop).
• Applied three convolutional neural network (CNN) architectures (ResNet, DenseNet, EfficientNet) and ensemble methods.

Main Results:

• DenseNet architectures outperformed ResNet and EfficientNet across both cropping methods.
• The ensemble model combining DenseNet121 from both cropping methods achieved an AUROC of 0.968, accuracy of 0.938, sensitivity of 0.980, and specificity of 0.903.
• Gradient-weighted class activation mapping indicated the model analyzes lymph node margins and adjacent soft tissue.

Conclusions:

• Deep learning models show significant promise for accurately detecting malignant ALNs in breast cancer.
• Integrating clinical considerations into image processing and using ensemble methods improves diagnostic precision.
• The developed model offers a potential tool to aid radiologists in breast cancer staging.