RcaNet:一种预测肿瘤突变负荷的深度学习模型

近年来，肺癌的发病率和死亡率不断攀升，已成为对人类生命健康威胁最大的恶性肿瘤之一，而非小细胞肺癌(NSCLC)发病率占肺癌总发病率的80%以上。由于其复杂的诊断过程和昂贵的诊断成本，NSCLC的有效诊断和治疗已成为医生面临的巨大挑战。医学相关研究发现，肿瘤突变负荷(TMB)与NSCLC免疫治疗的疗效呈正相关，同时TMB值对靶向治疗和化疗的疗效具有一定的预测作用。基于上述发现，本研究提出一种深度学习模型(RcaNet),该模型以残差网络(ResNet)为骨干网络，在网络内增加多维度特征注意和多尺度信息融合，以增强网络对肺癌病理组织切片深层特征的关注与提取能力。通过将RcaNet与主流深度学习模型在TCGA公开数据集上进行实验，实验训练样本数为925 954张。结果表明，RcaNet模型的性能平均曲线下面积(AUC)值为0.883 0,比现有结果最好的CAIM模型高出6.8%,比ResNeSt模型高出4.2%,比ResNet模型高出5.3%。研究结果对非小细胞肺癌诊断治疗有较强的指导意义，具有较高的应用价值。

RcaNet: A Deep Learning Model for Predicting Tumor Mutation Burden

In recent years, the morbidity and mortality of lung cancer have been rising continuously, and it has become one of the most dangerous malignant tumors that threaten human life and health. The incidence of non-small cell lung cancer (NSCLC) accounts for more than 80% of the total incidence of lung cancer. Due to its complicated diagnostic process and high diagnostic cost, the effective diagnosis and treatment of NSCLC have become a great challenge for doctors. It has found that tumor mutation burden (TMB) is positively correlated with the efficacy of NSCLC immunotherapy, and TMB value has a certain predictive effect on the efficacy of targeted therapy and chemotherapy. Based on the above findings, a deep learning model (RcaNet) was proposed. In this model, a residual network (ResNet) was taken as the backbone network, and multi-dimensional feature attention and multi-scale information fusion were added in the network, enhancing the ability of the network in paying attention to and extract the deep features of lung cancer pathological sections. Experiments were performed with RcaNet and the mainstream deep learning models on the TCGA public data set with experimental training samples of 925 954. The results showed that the average area under the curve (AUC) of the RcaNet model is 0.883 0, which is 6.8% higher than that of CAIM model, 4.2% higher than that of ResNeSt model, and 5.3% higher than that of ResNet model. Our proposed method has guiding significance and application value for the diagnosis and treatment of NSCLC.