Sun Shuhan, Chen Yajing, Liu Yutong, Li Cuiying, Miao Shumei, Yang Bin, Yu Feihong
Department of Ultrasound, The First Affiliated Hospital of Nanjing Medical University, Nanjing 210029, China.
Department of Information, The First Affiliated Hospital of Nanjing Medical University, Nanjing 210029, China.
Acad Radiol. 2025 Mar;32(3):1252-1263. doi: 10.1016/j.acra.2024.09.065. Epub 2024 Oct 15.
The aim of this study was to evaluate the capability of an ultrasound (US)-based deep learning (DL) nomogram for predicting axillary lymph node (ALN) status after neoadjuvant chemotherapy (NAC) in breast cancer patients and its potential to assist radiologists in diagnosis.
Two medical centers retrospectively recruited 535 node-positive breast cancer patients who had undergone NAC. Center 1 included 288 patients in the training cohort and 123 patients in the internal validation cohort, while center 2 enrolled 124 patients for the external validation cohort. Five DL models (ResNet 34, ResNet 50, VGG19, GoogLeNet, and DenseNet 121) were trained on pre- and post-NAC US images, and the best model was chosen. A US-based DL nomogram was constructed using DL predictive probabilities and clinicopathological characteristics. Furthermore, the performances of radiologists were compared with and without the assistance of the nomogram.
ResNet 50 performed best among all DL models, achieving areas under the curve (AUCs) of 0.837 and 0.850 in the internal and external validation cohorts, respectively. The US-based DL nomogram demonstrated strong predictive ability for ALN status post-NAC, with AUCs of 0.890 and 0.870 in the internal and external validation cohorts, respectively, outperforming both the clinical model and the DL model (p all < 0.05, except p = 0.19 for DL model in external validation cohort). Moreover, the nomogram significantly improved radiologists' diagnostic ability.
The US-based DL nomogram is promising for predicting ALN status post-NAC and could assist radiologists for better diagnostic performance.
本研究旨在评估基于超声（US）的深度学习（DL）列线图预测乳腺癌患者新辅助化疗（NAC）后腋窝淋巴结（ALN）状态的能力及其辅助放射科医生进行诊断的潜力。
两个医学中心回顾性招募了535例接受过NAC的淋巴结阳性乳腺癌患者。中心1的训练队列包括288例患者，内部验证队列包括123例患者，而中心2招募了124例患者作为外部验证队列。在NAC前和NAC后的US图像上训练了5种DL模型（ResNet 34、ResNet 50、VGG19、GoogLeNet和DenseNet 121），并选择了最佳模型。使用DL预测概率和临床病理特征构建了基于US的DL列线图。此外，比较了有和没有列线图辅助时放射科医生的表现。
ResNet 50在所有DL模型中表现最佳，在内部和外部验证队列中的曲线下面积（AUC）分别为0.837和0.850。基于US的DL列线图对NAC后ALN状态具有很强的预测能力，在内部和外部验证队列中的AUC分别为0.890和0.870，优于临床模型和DL模型（所有p值均<0.05，外部验证队列中DL模型的p值为0.19除外）。此外，列线图显著提高了放射科医生的诊断能力。
基于US的DL列线图在预测NAC后ALN状态方面很有前景，并且可以辅助放射科医生获得更好的诊断性能。
