基于全身PET-CT影像和多任务深度学习的食管癌淋巴结分期预测模型研究

在我国，食管癌发病率和死亡率分别居所有恶性肿瘤的第四位和第三位。尽管针对食管癌患者的诊断和治疗方法在愈加完善，但食管癌患者预后仍然较差。淋巴结转移是食管癌患者重要的独立预后危险因素。术前精准诊断淋巴结分期有助于个性化治疗方案的制定，尤其是新辅助放化疗的实施，对食管癌患者个体化精准治疗具有重要的临床意义。本项目拟基于全身PET-CT影像，利用多任务深度学习算法挖掘包含肿瘤、纵膈和腹腔在内的多个感兴趣区域中与淋巴结分期高度相关的影像特征，结合临床信息构建全自动的食管癌淋巴结分期术前预测模型，并在多中心前瞻性数据集上进行预测效能验证，最终辅助食管癌临床治疗决策。

The incidence and mortality of esophageal cancer (EC) rank the third and fourth in all malignancies in China. Although the diagnosis and treatment of EC have advanced, the prognosis of EC remains extremely poor. Lymph node (LN) metastasis is one of the most important independent prognostic factors, which generally indicates a worse outcome. Accurate preoperative LN staging is also important for making treatment decisions, such as neoadjuvant chemoradiotherapy, which would be of great clinical significance for EC. In this project, we will utilize the whole-body PET-CT images and multitask deep learning algorithm to build a fully automatic model for the preoperative prediction of LN staging by incorporating the clinical features and quantitative radiological features from multiple regions of interest including tumor, mediastinum and abdominal cavity. Moreover, we will build a multi-center and prospective cohort to validate the predictive model, aiming at making the model clinically applicable in assisting treatment decision making for EC patients.