Chen Chen, Liu Yuanzhen, Yao Jincao, Lv Lujiao, Pan Qianmeng, Wu Jinxin, Zheng Changfu, Wang Hui, Jiang Xianping, Wang Yifan, Xu Dong
Graduate School, Wannan Medical College, Wuhu, 241002, China.
Department of Diagnostic Ultrasound Imaging & Interventional Therapy, Zhejiang Cancer Hangzhou Institute of Medicine (HIM), Chinese Academy of Sciences, Hangzhou, 310022, China.
Heliyon. 2023 Aug 11;9(8):e19066. doi: 10.1016/j.heliyon.2023.e19066. eCollection 2023 Aug.
Both calcification and colloid in thyroid nodules are reflected as echogenic foci in ultrasound images. However, calcification and colloid have significantly different probabilities of malignancy. We explored the performance of a deep learning (DL) model in distinguishing the echogenic foci of thyroid nodules as calcification or colloid.
We conducted a retrospective study using ultrasound image sets. The DL model was trained and tested on 30,388 images of 1127 nodules. All nodules were pathologically confirmed. The area under the receiver-operator characteristic curve (AUC) was employed as the primary evaluation index.
The YoloV5 (You Only Look Once Version 5) transfer learning model for thyroid nodules based on DL detection showed that the average sensitivity, specificity, and accuracy of distinguishing echogenic foci in the test 1 group (n = 192) was 78.41%, 91.36%, and 77.81%, respectively. The average sensitivity, specificity, and accuracy of the three radiologists were 51.14%, 82.58%, and 61.29%, respectively. The average sensitivity, specificity, and accuracy of distinguishing small echogenic foci in the test 2 group (n = 58) was 70.17%, 77.14%, and 73.33%, respectively. Correspondingly, the average sensitivity, specificity, and accuracy of the radiologists were 57.69%, 63.29%, and 59.38%.
The study demonstrated that DL performed far better than radiologists in distinguishing echogenic foci of thyroid nodules as calcifications or colloid.
甲状腺结节中的钙化和胶体在超声图像中均表现为回声灶。然而，钙化和胶体的恶性概率存在显著差异。我们探讨了深度学习（DL）模型在区分甲状腺结节回声灶为钙化或胶体方面的性能。
我们使用超声图像集进行了一项回顾性研究。DL模型在1127个结节的30388张图像上进行训练和测试。所有结节均经病理证实。采用受试者操作特征曲线下面积（AUC）作为主要评估指标。
基于DL检测的甲状腺结节YoloV5（你只看一次版本5）迁移学习模型显示，在测试1组（n = 192）中区分回声灶的平均敏感性、特异性和准确性分别为78.41%、91.36%和77.81%。三位放射科医生的平均敏感性、特异性和准确性分别为51.14%、82.58%和61.29%。在测试2组（n = 58）中区分小回声灶的平均敏感性、特异性和准确性分别为70.17%、77.14%和73.33%。相应地，放射科医生的平均敏感性、特异性和准确性分别为57.69%、63.29%和59.38%。
该研究表明，在区分甲状腺结节回声灶为钙化或胶体方面，DL的表现远优于放射科医生。
