Imaging data help AI models predict lymph node metastasis
Deep learning models can be trained to predict lymph node metastasis in breast cancer patients, according to new findings published in Radiology.
“As the most common cancer among women worldwide, breast cancer poses a great challenge to public health on a global scale,” wrote Li-Qiang Zhou, Tongji Medical College in China, and colleagues. “Identification of the presence of lymph node metastasis is pivotal for the pathologic staging, prognosis and guidance of treatment in patients with breast cancer.”
The researchers gathered ultrasound (US) images from 756 breast cancer patients who received care at the same facility from 2016 to 2018. The team then trained three convolutional neural networks (CNNs) on 90% of the 2016-2018 data, testing them on the remaining 10% of that data (test set A) and an independent test set of images from another 78 breast cancer patients (test set B). The performance of these CNNs was compared to that of five radiologists with three to 26 years of experience.
Overall, the researchers found that all three CNNs achieved an area under the ROC curve (AUC) of at least 0.87 for predicting lymph node metastasis using US images from test set A. The best performance came from the team’s Inception V3 deep learning model, which had an AUC of 0.90, sensitivity of 82% and a specificity of 79%. For test set B, the CNNs all achieved an AUC of at least 0.86. The Inception V3 model once again turned in the strongest performance, achieving an AUC of 0.89, sensitivity of 85% and specificity of 72%.
The CNN models, the authors added, outperformed the radiologists “with a statistically significant difference.”
“Our results demonstrate the feasibility of using CNNs to predict whether early primary breast cancer will metastasize,” the authors wrote. “This work represents an improved approach to the assessment of early lymph node status based on the US images from patients with primary breast cancer obtained before surgery and significantly improves on current prediction methods that rely on physical examinations or lymph node imaging. To the best of our knowledge, this is the first study to apply the deep learning of CNNs for clinically negative lymph node metastasis prediction analysis.”
Michael has more than 18 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.