图/表详细信息

人工智能在胃癌诊疗和患者预后预测中的应用现状及未来展望

彭东阁, 万子叶, 卢宁

中国癌症杂志 2025, 35 (5): 496-504. DOI: 10.19401/j.cnki.1007-3639.2025.05.008

摘要（191）

HTML （2）

PDF（pc）（1042KB）（540）

胃癌是全球范围内高发的恶性肿瘤之一，具有起病隐匿、早期诊断困难、进展期预后不良等特点。传统诊断技术受主观因素影响较大，且在准确率和效率方面存在局限，难以满足精准医学的临床需求。近年来，人工智能（artificial intelligence，AI）技术，尤其是基于深度学习（deep learning，DL）的快速发展，为胃癌的精准诊疗带来了全新的机遇。AI辅助胃镜诊断可显著提升病变检出率及诊断效率，AI驱动的影像组学模型可精准预测肿瘤浸润深度、淋巴结及腹膜转移情况，而AI辅助病理学系统的应用则可以显著提高诊断的准确率和效率。此外，结合多组学数据的AI模型在化疗和靶向治疗反应预测以及个体化预后评估方面亦展现出巨大潜力。然而，AI技术在胃癌领域的临床转化仍面临诸多挑战，包括数据标准化不统一、模型泛化能力不足及算法可解释性较弱等问题。因此，本文系统综述AI技术在胃癌诊断、疗效评估及预后预测方面的最新研究进展，深入探讨当前技术所面临的核心挑战，并展望未来AI在胃癌精准诊疗中的发展趋势，以期推动AI技术的广泛应用和临床转化，最终实现胃癌诊疗的精准化和个体化，改善患者的临床预后。

Application	Research content	AI method	Results
Endoscopic diagnosis	Validation of AI model consistency with pathological diagnosis of gastric lesions^[16]	U-Net deep learning model	The AI model exhibited superior consistency with pathological diagnoses during real-time endoscopic video monitoring, surpassing endoscopists in diagnostic accuracy
	Effectiveness of AI-assisted real-time monitoring and diagnosis of endoscopic lesions ^[19]	CNN-based deep learning model	The AI-assisted system achieved an accuracy of 84.7%, sensitivity of 100%, and specificity of 84.3% in gastric cancer detection, enhancing the quality of endoscopic examinations
Imaging diagnosis	Prediction of preoperative lymph node metastasis in gastric cancer^[30]	DLRN based on CNN	DLRN effectively distinguished lymph node metastasis stages in advanced gastric cancer, outperforming clinical N staging methods and showing significant correlation with OS
	Prediction of occult peritoneal metastasis in advanced gastric cancer^[35]	GAN-based deep learning model (PMetNet)	The nomogram model, incorporating Lauren classification and tumor differentiation, significantly improved the diagnostic accuracy of preoperative occult peritoneal metastasis, with AUC values of 0.950 and 0.953, respectively
Pathological diagnosis	Development of an AI-assisted rapid pathological diagnosis system^[40]	CNN-based deep learning pathology diagnostic system	The system rapidly identified suspicious tumor regions, demonstrating robust diagnostic performance (AUC: 0.986; accuracy: 0.873; sensitivity: 0.996; specificity: 0.806)
	Identification of lymph nodes and tumor regions to assist in gastric cancer lymph node metastasis diagnosis^[42]	CNN-based deep reinforcement learning	The model achieved a sensitivity of 98.5% and specificity of 96.1%, significantly reducing the time required for pathologists to diagnose lymph node metastasis
Therapeutic response prediction	Prediction of targeted therapy efficacy in HER2-positive gastric cancer patients^[46]	Hybrid predictive model (MuMo) based on CNN and Transformer architecture	MuMo exhibited AUC values of 0.821 and 0.914 in predicting responses to targeted therapy in HER2-positive gastric cancer patients, effectively stratifying them into high-risk and low-risk groups
Prognostic assessment	Development of a RFS prediction model for advanced gastric cancer patients based on radiomic features^[54]	CNN-based deep learning model	The model optimized RFS prediction and high-risk stratification in gastric cancer patients

View table in article

表1 AI联合组学数据在胃癌中的研究进展

正文中引用本图/表的段落

随着人工智能（artificial intelligence，AI）在医疗领域的广泛应用，特别是在胃癌的诊断和预后预测方面，AI展现出巨大的潜力[3-4]。相比传统机器学习（machine learning，ML），深度学习（deep learning，DL）具有强大的特征学习能力、层次化表示和对大数据的适应性[5]。DL是ML的特殊类型，通过多层神经网络模拟人脑信息处理方式，能够自动从大量数据中学习特征，超越传统ML的表现（图1）。常用的DL模型包括深度神经网络、卷积神经网络（convolutional neural network，CNN）、图神经网络、残差网络、生成对抗网络及Transformer架构等[6]。本文将重点综述DL在胃癌诊断、疗效评估和预后预测中的应用现状及未来发展方向。

Zhou等[50]基于DL构建多模态注意力的虚拟多重免疫荧光（multiplexed immunofluorescence，mIF）染色系统来预测胃癌中多种与OS相关的生物标志物。该系统建立在端到端生成CNN之上，以自发荧光和4’,6-二脒基-2-苯基吲哚载玻片作为输入，生成胃癌相关生物标志物的mIF染色，结果显示，与CD3低表达组相比，CD3高表达组患者的OS显著更长（P <0.001），CD20、CD8、PD1和综合阳性评分与较长的OS显著相关，而CD163和FOXP3表现出较短OS的趋势。Cai等[51]通过DL建立了1 699例胃癌患者基因表达数据集的预后模型GPSGC，探索与GPSGC相关的生存调控机制和治疗靶点，结果显示，转化生长因子β1和血管内皮生长因子的基因集与GPSGC风险评分和不良预后显著相关，可作为胃癌患者的潜在治疗靶点。

AI与多组学数据的融合能够为胃癌诊疗研究提供多维度的分析框架。表1从应用场景、AI方法及核心成果3个维度，对不同技术路径下的研究进展进行归纳，旨在为将来多模态数据融合研究提供方向性参考。

GAN: Generative adversarial network; RFS: Recurrence-free survival. ...

Perioperative chemotherapy with fluorouracil plus leucovorin, oxaliplatin, and docetaxel versus fluorouracil or capecitabine plus cisplatin and epirubicin for locally advanced, resectable gastric or gastro-oesophageal junction adenocarcinoma (FLOT4): a randomised, phase 2/3 trial

2019

... 转化治疗已成为部分局部晚期胃癌患者的选择，但仅有20%~40%的患者能从新辅助化疗中获益[47].因此，精准识别潜在获益患者，避免不必要的治疗相关不良反应具有重要临床意义.Li等[48]基于CT影像中肿瘤及周围组织特征，构建了可有效预测Ⅱ/Ⅲ期胃癌患者DFS和化疗反应的影像组学模型.Tan等[49]则建立了delta影像组学模型，可有效预测晚期胃癌患者的化疗反应，最佳AUC达0.83.Gao等[31]开发的影像组学模型通过评估肿瘤中的调节性T细胞浸润情况，能够预测并识别在程序性死亡蛋白-1（programmed death-1，PD-1）单抗治疗中可能发生超进展的患者.尽管目前影像组学模型在化疗反应预测中表现优异，但仍缺乏特异性生物标志物来精确识别获益患者，应当继续探索能够与影像学特征相结合的生物标志物. ...

Intratumoral and peritumoral radiomics of contrast-enhanced CT for prediction of disease-free survival and chemotherapy response in stage Ⅱ/Ⅲ gastric cancer

2020

Predicting chemotherapeutic response for far-advanced gastric cancer by radiomics with deep learning semi-automatic segmentation

2020

Virtual multiplexed immunofluorescence staining from non-antibody-stained fluorescence imaging for gastric cancer prognosis

2024

... Zhou等[50]基于DL构建多模态注意力的虚拟多重免疫荧光（multiplexed immunofluorescence，mIF）染色系统来预测胃癌中多种与OS相关的生物标志物.该系统建立在端到端生成CNN之上，以自发荧光和4’,6-二脒基-2-苯基吲哚载玻片作为输入，生成胃癌相关生物标志物的mIF染色，结果显示，与CD3低表达组相比，CD3高表达组患者的OS显著更长（P <0.001），CD20、CD8、PD1和综合阳性评分与较长的OS显著相关，而CD163和FOXP3表现出较短OS的趋势.Cai等[51]通过DL建立了1 699例胃癌患者基因表达数据集的预后模型GPSGC，探索与GPSGC相关的生存调控机制和治疗靶点，结果显示，转化生长因子β1和血管内皮生长因子的基因集与GPSGC风险评分和不良预后显著相关，可作为胃癌患者的潜在治疗靶点. ...

Identification of a tumor microenvironment-relevant gene set-based prognostic signature and related therapy targets in gastric cancer

2020

Gastric cancer, version 2.2022, NCCN clinical practice guidelines in oncology

2022

... HER2在多种组织（包括胃肠道）中表达，能够促进细胞增殖并抑制凋亡.有研究[52]表明，HER2过表达是胃癌进展的重要因素，并且与患者预后不良相关.刘洋等[41]用AI辅助显微镜评估新辅助化疗（neoadjuvant chemotherapy，NAC）前后胃癌病理学标本，结果显示，在非病理学完全缓解（pathologic complete response，pCR）的患者中，NAC后HER2表达下调者的5年OS率和DFS率均高于上调者（56.7% vs 30.5%，P<0.001；56.1% vs 23.0%，P<0.001）；在非pCR的HER2 2+/3+患者中，表达下调者的5年OS率和DFS率优于表达无变化者（56.7% vs 33.5%，P=0.003；56.1% vs 32.0%，P=0.002）.此外，Ma等[53]基于门脉期增强CT的影像组学成功预测HER2高表达，训练集和验证集的AUC分别为0.84和0.78.尽管AI模型在评估一致性方面表现出色，但由于HER2表达的异质性，AI可能无法完全捕捉到不同患者的个体差异. ...

A multiphase contrast-enhanced CT radiomics model for prediction of human epidermal growth factor receptor 2 status in advanced gastric cancer

2022

A radiomics-based model for predicting prognosis of locally advanced gastric cancer in the preoperative setting

2021

... Research progress of AI combined with omics data in gastric cancer

本文的其它图/表