Automatic assessment of axillary lymph node status in cT1-2N0 Breast Cancer with US modality-adaptive network
Tuesday, April 9, 2024
12:14pm – 12:21pm
Location: 412
Authors: Yuanjing Gao, Peking Union Medical Hospital
This study aimed to develop a reliable preoperative diagnostic method for axillary lymph node (ALN) metastasis in early breast cancer using a deep learning model based on axillary contrast-enhanced ultrasound (CEUS). The modality-adaptive network (MAN) was introduced, capable of processing either grayscale or color Doppler ultrasound images to determine the probability of heavy ALN tumor burden. In a cohort of 275 patients, the MAN demonstrated a high diagnostic performance with an area under the receiver operating characteristic curve (AUC) of 0.91 to 0.98. Additionally, when combined with automated lesion detection technology, the model achieved an AUC of 0.82 on an independent test set. Notably, this approach outperformed existing AI models, providing a valuable tool for patients with multi-focal lesions or those undergoing primary lesion therapy. The MAN, in conjunction with clinical parameters, offers a direct and efficient preoperative assessment of ALN involvement, potentially reducing the need for unnecessary procedures in certain patient groups.