分子靶向治疗已成为肺癌临床标准治疗的重要内容，但及时、无创、便捷、精准地预测靶向药物疗效是亟需解决的关键问题。细胞代谢异常作为肿瘤高特征性标志，是靶向药物疗效评价的有效指证。以多模态代谢影像为核心并包括代谢组学、代谢酶相关蛋白组学、转录组学等多维组学数据，不仅涵盖了从宏观到微观、从表型到机理的完整信息，而且以代谢为纽带形成紧密的逻辑链，成为靶向药物评价的关键切入点。为此，本课题组提出“定量代谢影像及代谢组学等多维数据关联互动式深度学习，人工智能评价肿瘤表型和预测靶向干预效果模型”。拟首先通过建立肺癌靶向药物干预动物模型，获取肺癌靶向耐药相关代谢数据；其次，基于自编码器模型和深度迁移学习框架，研究基于代谢影像组学数据互动的特征提取和分析方法、高通量数据比对与并行算法等关键技术；最后通过多中心、大样本临床研究验证肺癌靶向耐药智能预测模型及其临床价值，以期提高肺癌靶向药物评价的精准性和预测性。

Targeted therapy has become an essential part of the clinical treatment of lung cancer. Namely, non-invasive, convenient and accurate prediction of the efficacy of targeted drugs is an important clinical issues. Dysregulation of cell metabolism, as a hallmark of cancer, is an effective method of assessing the efficacy of targeted drugs. Albeit composed of multimodal metabolic imaging and metabolic enzyme-related proteomics, transcriptomics and other omics, the complex metabolic data, indicates several clinical pathways from macro to micro, from phenotype to mechanism, and the response chain through metabolism. Thanks to recent advances in assessment of metabolism-dependent drug efficacy, a deep learning technique is proposed and validated to predict the effect of targeted drugs, in terms of quantitative metabolic imaging, metabolomics and other omics. First, a multidimensional metabolic data set is constructed upon a lung cancer model. Then, several machine learning techniques including feature extraction, alignment and selection are studied using the autoencoder and transfer learning neural network. Finally, a multicentre, large-sample clinical study is conducted to validate the predictive model of lung cancer-targeted drugs. We envision the outcome of the underlying pioneer study as an emerging tool in assessment of lung cancer-targeted drugs.