切换至 "中华医学电子期刊资源库"

中华医学超声杂志(电子版) ›› 2026, Vol. 23 ›› Issue (01) : 1 -7. doi: 10.3877/cma.j.issn.1672-6448.2026.01.001

专家论坛

基于深度学习的超声影像在儿童甲状腺结节风险分层中的应用进展
丁娇娇, 石文媛()   
  1. 100045 国家儿童医学中心(北京)首都医科大学附属北京儿童医院介入医学科
  • 收稿日期:2026-01-12 出版日期:2026-01-01
  • 通信作者: 石文媛
  • 基金资助:
    儿童创新医疗器械概念验证平台(首都医科大学附属北京儿童医院)(2024030119)

Advances in deep learning-based ultrasound imaging for risk stratification of pediatric thyroid nodules

Jiaojiao Ding, Wenyuan Shi()   

  • Received:2026-01-12 Published:2026-01-01
  • Corresponding author: Wenyuan Shi
引用本文:

丁娇娇, 石文媛. 基于深度学习的超声影像在儿童甲状腺结节风险分层中的应用进展[J/OL]. 中华医学超声杂志(电子版), 2026, 23(01): 1-7.

Jiaojiao Ding, Wenyuan Shi. Advances in deep learning-based ultrasound imaging for risk stratification of pediatric thyroid nodules[J/OL]. Chinese Journal of Medical Ultrasound (Electronic Edition), 2026, 23(01): 1-7.

表1 CNN、YOLO 与 Vision Transformer 在儿科甲状腺超声诊断中的任务定位与场景化选择
技术架构 任务定位 优点 局限性 儿童优先适用场景
CNN 图像分类、结合U-Net等做分割 对局部纹理、边缘形态特征学习稳定;小样本下迁移学习效果好 对“全局上下文/长程依赖”建模不足;跨设备/跨中心分布偏移时性能易波动;对腺体背景整合能力受限;若训练集病例谱偏“高危富集”,可能出现“高敏感-低特异” 离线分析优先:门诊/住院后处理;单中心或小规模多中心较易落地;可作为“风险分层概率输出 + 医师复核”的基础模型
YOLO 目标检测(框出结节、同步给出良恶性或可疑度) 单次前向传播同时完成定位与分类,强调高效率;更适配动态扫查(视频流/实时提示) 若只基于单帧,仍可能受操作者取帧影响;对标注质量与框一致性敏感 实时扫查优先:儿科配合度低、检查时间短,适合“实时提示可疑结节”与“减少漏扫”
ViT 多视图融合、全局上下文建模;可扩展到“时空序列/多模态” 自注意力更擅长整合全局信息与多视角信息;对“小结节 + 多切面信息融合”的理论契合度更高;在多中心数据上可与域泛化策略结合 对数据规模与多样性更敏感,小样本条件下易不稳定;跨中心域偏移仍需专门策略;工程化与端侧部署门槛相对更高 离线精细分层优先:尤其是“横切+纵切”多视图输入、需要综合背景腺体信息的场景
表2 不同研究的人工智能模型在儿童甲状腺结节风险分层中的性能对比
研究 人群 临床应用场景 评价指标 验证层级 局限性
Richman等[3] 儿童队列 成人RSS在儿科表现 ACR TI-RADS:22.1%(17/77)的恶性结节在初诊时不会被推荐细针穿刺活检 L1/L2 单中心数据;规则系统中等效能反映了成人阈值在儿科的不适配;未针对儿科特异性背景(如AIT)进行分层分析
Yang等[4] 139例,≤21岁 结节良恶性诊断:成人训练深度学习模型 vs ACR TI-RADS vs 医师印象 深度学习模型:敏感度 87.5%(95%置信区间:78.3~95.5),特异度 36.1%(95%置信区间:25.6~46.8) L1 单中心回顾性设计;样本量相对有限(n=139);特异度仅36.1%,可能导致不必要穿刺活检增加;未评估跨中心、跨设备的分布偏移
Ha等[14] 儿童多队列 结节良恶性诊断:成人训练甲状腺AI诊断模型外部测试 横切面AUC 0.913;纵切面AUC 0.929;横纵切面联合AUC 0.927 L3 虽为多中心外部验证,但模型训练基础来自成人数据,儿科特异性病理亚型(如弥漫性硬化型)的覆盖程度未明确;模型可解释性未涉及
Kim等[15] 儿童队列 5种成人RSS穿刺标准比较 以穿刺标准可接受且可通过调整阈值改善为主 L1/L2 样本量与病理亚型分布未详细披露;结节大小与最高危分级的儿科化改进方案仍缺乏前瞻性验证
Zhang等[16] 54 305例 针对报告文本,采用自然语言处理,尤其是长短期记忆网络等 内部测试AUC 0.965;外部测试AUC 0.912 L3 虽样本量大,但基于文本报告而非原始图像,易含结论性信息导致数据泄漏风险;儿科患者占比与儿科特异性表述习惯的适用性需专门验证
Li等[17] 儿童PTC队列 甲状腺外侵犯预测:超声放射组学+LightGBM 内部测试AUC 0.787;外部测试AUC 0.770 L3 终点为甲状腺外侵犯预测而非良恶性鉴别,临床应用场景相对专科化;特征提取的可重复性与跨设备稳定性未充分评估;阈值的临床转化路径需进一步明确
Qian等[18] 儿童PTC队列 颈部淋巴结转移预测 多模态列线图AUC 0.85~0.90 L2/L3 多中心验证数据有限;临床阈值设定需进一步优化
Ni等[19] 不确定细胞学结节 Bethesda Ⅲ/Ⅳ类结节风险分层 AI+分子标记物模型AUC 0.88 L2 样本量相对较小;儿科特异性队列数据不足
1
国家儿童医学中心, 国家儿童肿瘤监测中心, 中华医学会小儿外科学分会, 等. 中国儿童甲状腺结节及分化型甲状腺癌专家共识[J]. 中华实用儿科临床杂志, 2020, 35(20): 1521-1530.
2
Xing Z, Qiu Y, Zhu J, et al. Diagnostic performance of adult-based ultrasound risk stratification systems in pediatric thyroid nodules: a systematic review and meta-analysis[J]. Front Endocrinol (Lausanne), 2023, 14: 1187935.
3
Richman DM, Benson CB, Doubilet PM, et al. Assessment of American College of Radiology Thyroid Imaging Reporting and Data System (TI-RADS) for pediatric thyroid nodules[J]. Radiology, 2020, 294(2): 415-420.
4
Yang J, Page LC, Wagner L, et al. Thyroid nodules on ultrasound in children and young adults: comparison of diagnostic performance of radiologists' impressions, ACR TI-RADS, and a deep learning algorithm[J]. AJR Am J Roentgenol, 2023, 220(3): 408-417.
5
Bulanova Pekova B, Sykorova V, Mastnikova K, et al. RET fusion genes in pediatric and adult thyroid carcinomas: cohort characteristics and prognosis[J]. Endocr Relat Cancer, 2023, 30(12): e230117.
6
Szwarkowska M, Kaleta K, Jurek A, et al. Occult thyroid cancer in autoimmune thyroiditis: lymph node ultrasound as the sole diagnostic indicator of malignancy in a pediatric case of papillary thyroid carcinoma[J]. Children (Basel), 2025, 12(2): 194.
7
Ma J, Zhang L, Sun Y, et al. Evaluating modified ACR TI-RADS nodule size for thyroid nodules in pediatric population[J]. BMC Med Imaging, 2025, 25(1): 269.
8
Li Y, Li T, He K, et al. A predictive nomogram of thyroid nodules based on deep learning ultrasound image analysis[J]. Front Endocrinol (Lausanne), 2025, 16: 1504412.
9
Liang XW, Cai YY, Yu JS, et al. Update on thyroid ultrasound: a narrative review from diagnostic criteria to artificial intelligence techniques[J]. Chin Med J (Engl), 2019, 132(16): 1974-1982.
10
Yang J. Machine learning methods for the analysis of thyroid nodule ultrasound images[D]. Durham, North Carolina: Duke University, 2024.
11
Wildman-Tobriner B, Taghi-Zadeh E, Mazurowski MA. Artificial intelligence (AI) tools for thyroid nodules on ultrasound, from the AJR special series on AI applications[J]. AJR Am J Roentgenol, 2022, 219(4): 547-554.
12
Yang J, Luo Z, Wen Y, et al. Artificial intelligence-enhanced ultrasound imaging for thyroid nodule detection and malignancy classification: a study on YOLOv11[J]. Quant Imaging Med Surg, 2025, 15(9): 7964-7976.
13
Feng N, Zhao S, Wang K, et al. Deep learning model for diagnosis of thyroid nodules with size less than 1 cm: A multicenter, retrospective study[J]. Eur J Radiol Open, 2024, 13: 100609.
14
Ha EJ, Lee JH, Mak N, et al. A deep learning-based artificial intelligence model assisting thyroid nodule diagnosis and management: pilot results for evaluating thyroid malignancy in pediatric cohorts[J]. Thyroid, 2025, 35(6): 652-661.
15
Kim PH, Yoon HM, Baek JH, et al. Diagnostic performance of five adult-based US risk stratification systems in pediatric thyroid nodules[J]. Radiology, 2022, 305(1): 190-198.
16
Zhang Q, Zhang S, Li J, et al. Improved diagnosis of thyroid cancer aided with deep learning applied to sonographic text reports: a retrospective, multi-cohort, diagnostic study[J]. Cancer Biol Med, 2022, 19(5): 733-741.
17
Li J, Xia F, Wang X, et al. Multiclassifier radiomics analysis of ultrasound for prediction of extrathyroidal extension in papillary thyroid carcinoma in children[J]. Int J Med Sci, 2023, 20(2): 278-286.
18
Qian J, Zhang Z, Chen Y, et al. Multimodal ultrasonographic and clinicopathological model for predicting high-volume lymph node metastasis in cN0 papillary thyroid carcinoma[J]. Front Endocrinol(Lausanne), 2025, 16: 1613672.
19
Ni JH, Liu YY, Chen C, et al. Optimizing thyroid nodule management with artificial intelligence: multicenter retrospective study on reducing unnecessary fine needle aspirations[J]. JMIR Med Inform, 2025, 13: e71740.
20
Richman DM, Cherella CE, Smith JR, et al. Clinical utility of sonographic features in indeterminate pediatric thyroid nodules[J]. Eur J Endocrinol, 2021, 184(5): 657-665.
21
Lee H, Chai YJ, Joo H, et al. Federated learning for thyroid ultrasound image analysis to protect personal information: Validation study in a real health care environment[J]. JMIR Med Inform, 2021, 9(5): e25869.
22
彭馨雨, 刘远征. 大数据时代儿童的个人信息保护制度的完善[J]. 争议解决, 2024, 10: 331-338.
23
潘燕杰. 我国健康医疗数据治理的标准化路径——以美国相关法律为借鉴[J]. 南京医科大学学报(社会科学版), 2023(5): 423-429.
24
Zhou Y, Chen C, Yao J, et al. A deep learning based ultrasound diagnostic tool driven by 3D visualization of thyroid nodules[J]. NPJ Digit Med, 2025, 8(1): 126.
No related articles found!
阅读次数
全文


摘要


AI


AI小编
你好!我是《中华医学电子期刊资源库》AI小编,有什么可以帮您的吗?