人工智能辅助多模态超声对甲状腺微小癌淋巴结转移的预警及机制研究

Thyroid cancer is one of the fastest growing malignant tumors in recent years. It has become the fifth most common malignant tumor in China, which is mainly caused by papillary thyroid carcinoma (PTC). Due to the improvement of high frequency ultrasound resolution, more and more thyroid cancer less than or equal to 1 cm could be found, which has been more than 50% of the thyroid carcinoma.Reports of papillary thyroid carcinoma cervical lymph node metastasis rate is as high as 10% to 60%, mainly for central lymph node metastasis, preoperative diagnostic sensitivity is only 10.9% ~ 42.0%.There is no warning and clear check means to predict early lymph node metastasis , so the majority of patients with passive accepted a broaden operation, leading to parathyroid and laryngeal nerve injury and seriously affect the quality of life of patients. Therefore this topic intends to use multimodal ultrasound combined with artificial intelligence based computer assisted imaging quantitative analysis technology, and with tissue pathology and molecular pathology, explore the ultrasonographic features of papillary thyroid microcarcinoma with or without lymph node metastasis, establish artificial intelligence based computer aided detected method using ultrasound imaging of thyroid microcarcinoma to predict early lymph node metastasis, improve the objectivity and accuracy of ultrasound diagnosis, make patients avoid from excessive surgical treatment , and the patients who have high risk of thyroid cancer metastasis get early diagnosis and treatment.

甲状腺癌是近年来全球发病率增长最快的恶性肿瘤，已上升为我国女性恶性肿瘤发病率第五位，其中50%以上为1cm以下的甲状腺微小乳头状癌（PTMC）。文献报告甲状腺微小癌颈部淋巴结的转移率10%～60%，主要转移至气管旁中央区淋巴结，因气管遮挡，作为首选的超声术前诊断敏感性仅为10.9%～42.0%。由于甲状腺微小癌淋巴结转移目前尚无有效的检查手段，因此多数患者被动接受了甲状腺全切或大部切除加颈部淋巴结清扫的扩大手术，因此2%-4%的患者导致了甲状旁腺或喉返神经损伤而严重影响了生活质量。本课题拟运用多模态超声技术结合计算机人工智能辅助成像定量分析技术，与组织病理和分子病理对照研究，探索有或无淋巴结转移的甲状腺微小癌的超声多模态综合影像特征及其分子机制，建立人工智能辅助多模态超声对甲状腺微小癌淋巴结转移的预警系统，以提高超声诊断的客观性和准确性，使部分患者免于扩大手术的风险。

甲状腺癌术后颈部淋巴结转移发生率高达35%，超声检查对中央区淋巴结转移的敏感度较低，仅为10.9%~36.2%，预防性中央区淋巴结清扫术会增加手术并发症的发生率，因此，精准预测中央区淋巴结转移对选择最佳手术方式和预后至关重要。针对上述背景，本研究拟通过建立并训练甲状腺癌淋巴结转移人工智能辅助成像预警系统，以提高甲状腺癌及颈部淋巴结转移的诊断准确率，为临床决策提供更多参考依据。主要研究内容包括：1.建立甲状腺结节多模态超声特征的原始数据库及甲状腺结节人工智能辅助超声诊断系统，通过多中心研究验证并完善该系统的诊断效能；2.建立甲状腺癌淋巴结转移超声影像数据库并构建标注软件和图像语义分割网络，最终构建预测模型并对其诊断效能进行训练和验证。通过获取超声多模态图像、术后组织病理与分子病理、临床资料等关键数据，取得以下两项重要结果：1、甲状腺结节人工智能辅助超声诊断系统诊断甲状腺癌表现良好，内部测试集中曲线下面积（AUC）值为0.911，外部测试集AUC值为0.881，其诊断效能不亚于高年资临床医生，并有助于提高低年资医生诊断效能；2、甲状腺癌颈部中央区淋巴结转移预测模型诊断甲状腺癌颈部中央区淋巴结转移状态表现良好，在内部测试集中预测中央区淋巴结转移的准确性、敏感性、特异性、阳性预测值和阴性预测值分别为0.80，0.76，0.83，0.80和0.79，AUC可达0.794。本项研究结果将有利于提高低年资医师对甲状腺癌及颈部淋巴结转移的诊断准确率，改善甲状腺结节良恶性受超声医师经验影响大、术前超声评估甲状腺癌颈部中央区淋巴结转移敏感度低的临床现状，为临床决策提供更多有力参考数据。