Department of Ultrasound, Shengjing Hospital of China Medical University, Shenyang, China.
Front Endocrinol (Lausanne). 2024 Apr 3;15:1299686. doi: 10.3389/fendo.2024.1299686. eCollection 2024.
To apply machine learning to extract radiomics features from thyroid two-dimensional ultrasound (2D-US) combined with contrast-enhanced ultrasound (CEUS) images to classify and predict benign and malignant thyroid nodules, classified according to the Chinese version of the thyroid imaging reporting and data system (C-TIRADS) as category 4.
This retrospective study included 313 pathologically diagnosed thyroid nodules (203 malignant and 110 benign). Two 2D-US images and five CEUS key frames ("2 second after the arrival time" frame, "time to peak" frame, "2 second after peak" frame, "first-flash" frame, and "second-flash" frame) were selected to manually label the region of interest using the "Labelme" tool. A total of 7 images of each nodule and their annotates were imported into the Darwin Research Platform for radiomics analysis. The datasets were randomly split into training and test cohorts in a 9:1 ratio. Six classifiers, namely, support vector machine, logistic regression, decision tree, random forest (RF), gradient boosting decision tree and extreme gradient boosting, were used to construct and test the models. Performance was evaluated using a receiver operating characteristic curve analysis. The area under the curve (AUC), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy (ACC), and F1-score were calculated. One junior radiologist and one senior radiologist reviewed the 2D-US image and CEUS videos of each nodule and made a diagnosis. We then compared their AUC and ACC with those of our best model.
The AUC of the diagnosis of US, CEUS and US combined CEUS by junior radiologist and senior radiologist were 0.755, 0.750, 0.784, 0.800, 0.873, 0.890, respectively. The RF classifier performed better than the other five, with an AUC of 1 for the training cohort and 0.94 (95% confidence interval 0.88-1) for the test cohort. The sensitivity, specificity, accuracy, PPV, NPV, and F1-score of the RF model in the test cohort were 0.82, 0.93, 0.90, 0.85, 0.92, and 0.84, respectively. The RF model with 2D-US combined with CEUS key frames achieved equivalent performance as the senior radiologist (AUC: 0.94 vs. 0.92, = 0.798; ACC: 0.90 vs. 0.92) and outperformed the junior radiologist (AUC: 0.94 vs. 0.80, = 0.039, ACC: 0.90 vs. 0.81) in the test cohort.
Our model, based on 2D-US and CEUS key frames radiomics features, had good diagnostic efficacy for thyroid nodules, which are classified as C-TIRADS 4. It shows promising potential in assisting less experienced junior radiologists.
应用机器学习从甲状腺二维超声（2D-US）联合对比增强超声（CEUS）图像中提取影像组学特征，对根据中文版甲状腺影像报告和数据系统（C-TIRADS）分类为 4 类的良性和恶性甲状腺结节进行分类和预测。
本回顾性研究纳入了 313 例经病理诊断的甲状腺结节（203 例恶性和 110 例良性）。选择 2 个 2D-US 图像和 5 个 CEUS 关键帧（“到达时间后 2 秒”帧、“达峰时间”帧、“达峰后 2 秒”帧、“首闪”帧和“次闪”帧），使用“Labelme”工具手动标记感兴趣区域。每个结节的 7 张图像及其标注均被导入 Darwin Research Platform 进行影像组学分析。数据集以 9:1 的比例随机分为训练集和测试集。使用支持向量机、逻辑回归、决策树、随机森林（RF）、梯度提升决策树和极端梯度提升等 6 种分类器构建和测试模型。使用受试者工作特征曲线分析评估性能。计算曲线下面积（AUC）、敏感度、特异度、阳性预测值（PPV）、阴性预测值（NPV）、准确性（ACC）和 F1 评分。一位初级放射科医生和一位高级放射科医生回顾了每个结节的 2D-US 图像和 CEUS 视频，并做出诊断。然后，我们将他们的 AUC 和 ACC 与我们最佳模型的 AUC 和 ACC 进行了比较。
初级放射科医生和高级放射科医生诊断 US、CEUS 和 US 联合 CEUS 的 AUC 分别为 0.755、0.750、0.784、0.800、0.873、0.890。RF 分类器的表现优于其他五种，其在训练集的 AUC 为 1，在测试集的 AUC 为 0.94（95%置信区间 0.88-1）。RF 模型在测试集的灵敏度、特异度、准确性、PPV、NPV 和 F1 评分分别为 0.82、0.93、0.90、0.85、0.92 和 0.84。RF 模型结合 2D-US 和 CEUS 关键帧的表现与高级放射科医生相当（AUC：0.94 与 0.92， = 0.798；ACC：0.90 与 0.92），且优于初级放射科医生（AUC：0.94 与 0.80， = 0.039，ACC：0.90 与 0.81）。
我们的模型基于 2D-US 和 CEUS 关键帧的影像组学特征，对 C-TIRADS 4 类甲状腺结节具有良好的诊断效能。它在辅助经验较少的初级放射科医生方面具有很大的潜力。
