基于病理组织切片的肺腺癌肿瘤突变负荷预测模型  

Prediction Model of Tumor Mutation Burden for Lung Adenocarcinoma Based on Pathological Tissue Slice

作　　者：孟祥福[1] 杨子毅 杨啸林[2] 侯嘉玥 Meng Xiangfu;Yang Ziyi;Yang Xiaolin;Hou Jiayue(School of Electronics and Information Engineering,Liaoning Technical University,Huludao 125000,Liaoning,China;Institute of Basic Medical Sciences,Chinese Academy of Medical Sciences,School of Basic Medicine,Peking Union Medical College,Beijing 100005,China;School of Communications and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]辽宁工程技术大学电子与信息工程学院,辽宁葫芦岛125000 [2]中国医学科学院基础医学研究所,北京协和医学院基础学院,北京100005 [3]南京邮电大学通信与信息工程学院,南京210023

出　　处：《中国生物医学工程学报》2023年第6期698-709,共12页Chinese Journal of Biomedical Engineering

基　　金：国家自然科学基金(61772249);中国医学科学院医学与健康科技创新工程项目(2018-I2M-AI-009);辽宁省教育厅一般项目(LJKZ0355)。

摘　　要：肺癌是目前死亡率最高的恶性癌症之一,其中非小细胞肺癌(NSCLC)致死率极高。最近医学研究发现,肿瘤突变负荷(TMB)对于癌症的免疫治疗和化疗的疗效具有较好的预测作用,但传统使用基因测序计算TMB的方法存在检测成本高、周期长、样本依赖度高等缺点。针对上述问题,本研究提出一种混合卷积神经网络和自注意力机制的深度学习模型(FCA-Former)用于预测TMB。该模型以CoAtNet为骨干网络,通过在网络中结合坐标注意力以及融合深度可分离卷积的方式,提高模型的运算速度以及对病理组织切片图像的全局特征提取能力。实验数据采用TCGA数据库中肺腺癌数字病理切片图像数据集,其中高TMB水平的样本271张,低TMB水平的样本66张。实验结果表明,所提方法达到的最高平均曲线下面积(AUC)为98.1%,比现有最好方法RcaNetr提高9.8%。此项研究结果对于NSCLC的预后治疗效果具有较强的指导意义。Lung cancer is one of the deadliest malignancies,particularly non-small cell lung cancer(NSCLC),poses a significant threat to public health.Recent medical research has found out the crucial role of tumor mutation burden(TMB)in predicting the efficacy of immunotherapy and chemotherapy for cancer treatment.However,traditional methods for calculating TMB through genetic sequencing suffer from drawbacks,such as high detection costs,lengthy processing periods,and sample dependency.To address above problems,this paper proposed a novel deep learning model named as FCA-Former,which combined convolutional neural networks and self-attention mechanisms to predict TMB.The model employed CoAtNet as a backbone network,integrating coordinate attention and depth wise separable convolutions to enhance computational efficiency and global feature extraction capabilities from pathological tissue biopsy images.Experimental data sourced from the TCGA database comprised a dataset of lung adenocarcinoma digital pathology images,including 271 samples with high TMB levels and 66 samples with low TMB levels.The experimental results demonstrated the effectiveness of the proposed approach,achieving a remarkable maximum area under the curve(AUC)of 98.1%.This AUC outperformed the state-of-the-art RcaNet method by 9.8%.The results of this study have significant implications for guiding prognostic and therapeutic strategies for NSCLC patients.

关 键 词：非小细胞肺癌 肿瘤突变负荷(TMB) 卷积神经网络 

分 类 号：R318[医药卫生—生物医学工程]