Axillary Lymph Node Evaluation Utilizing Convolutional Neural Networks Using MRI Dataset.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Axillary lymph node metastasis is a critical prognostic factor in breast cancer.
• Accurate prediction of metastasis is essential for treatment planning.
• Current diagnostic methods can be invasive, carrying potential risks and costs.

Purpose of the Study:

• To evaluate the efficacy of a convolutional neural network (CNN) for predicting axillary lymph node metastasis using breast MRI data.
• To assess the feasibility of deep learning models in identifying metastatic lymph nodes non-invasively.

Main Methods:

• Retrospective analysis of 275 axillary lymph nodes from breast MRI scans (2013-2016).
• 3D segmentation of lymph nodes followed by extraction of 32x32 image patches.
• Development and training of a seven-layer CNN with data augmentation and regularization using TensorFlow.
• Evaluation of the CNN model using five-fold cross-validation.

Main Results:

• The CNN model achieved a mean five-fold cross-validation accuracy of 84.3% in predicting axillary lymph node metastasis.
• The model demonstrated feasibility in classifying lymph nodes as metastatic or non-metastatic.
• Softmax score threshold of 0.5 was used for classification.

Conclusions:

• Deep convolutional neural network architectures are capable of predicting the likelihood of axillary lymph node metastasis from breast MRI.
• The developed CNN model shows potential as a non-invasive alternative to core needle biopsy and sentinel lymph node evaluation.
• Expansion of the dataset is expected to further enhance the prediction model's performance.