Deep learning-powered multi-parametric ultrasound for classifying metastatic versus reactive axillary lymph nodes

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate classification of axillary lymph nodes (ALNs) is crucial for breast cancer staging and treatment decisions.
• Current diagnostic methods for ALN metastasis have limitations, necessitating improved imaging techniques.
• Deep learning offers potential for enhancing the accuracy of ALN metastasis detection.

Purpose of the Study:

• To develop and validate a multi-parametric ultrasound imaging-based deep learning method for classifying metastatic versus non-metastatic ALNs in breast cancer patients.
• To integrate conventional ultrasound B-mode, shear wave elastography, and color Doppler imaging for enhanced classification.
• To evaluate the performance of a novel deep learning network for ALN classification.

Main Methods:

• A transfer learning approach using a pretrained MobileNetv2 network with a custom shallow head was employed.
• The network incorporated convolutional neural networks with mixed pooling, weighted sum mixed pooling, and squeeze-and-excite attention mechanisms.
• The method was trained and validated on data from 174 breast cancer patients using five-fold cross-validation.

Main Results:

• The proposed multi-parametric deep learning method achieved a mean classification accuracy of 0.91 and a cross-validated AUC (cvAUC) of 0.92.
• The model demonstrated high sensitivity (0.93) and specificity (0.91), outperforming existing state-of-the-art deep learning models for ALN classification.
• Ablation studies confirmed model robustness, and multi-parametric imaging significantly improved performance and reduced confidence interval width compared to uni-parametric data.

Conclusions:

• Integration of multi-parametric ultrasound imaging with deep learning significantly enhances the classification of metastatic and non-metastatic ALNs.
• The developed method offers a robust and reliable tool for improving breast cancer diagnosis and patient management.
• This approach holds promise for more accurate staging and personalized treatment strategies in breast cancer care.